HiPEACinfo 64

Page 1

64 OCTOBER 2021

ng i t u p m o C Systems Week Lyon

Alfonso Valencia on the future of medicine Powering biomedical breakthroughs with HiPEAC tech Why now is the time to get into quantum Silexit strategy: From spinoff to acquisition


contents

Bienvenue à Lyon

3

4

Alfonso Valencia on biological digital twins

Welcome Koen De Bosschere

4 News 10 HiPEAC voices ‘For the first time, we have real biological data to create digital twins’ Alfonso Valencia 12 Healthcare and biomedicine special feature First aid: How HiPEAC technologies are powering medical breakthroughs Martin Žagar, Horacio Pérez-Sánchez, Arnau Montagud, José Carbonell, Camilla Giunti, Juan Mompeán, Juan Luis Aragón, Pablo Artal and Marius Rus 19 Technology transfer Silexit strategy: How Silexica went from academic spinoff to acquisition by Xilinx Rainer Leupers and Maximilian Odendahl 22 Innovation booster EU services to support your innovation pathways Xavier Salazar 24 Peac performance Speeding up simulations for drug discovery with Autodock-GPU Leonardo Solis-Vasquez, Andreas F. Tillack, Diogo SantosMartins, Andreas Koch and Stefano Forli 26 SME snapshot Fast-tracking pharmaceuticals with Nostrum Biodiscovery Ezequiel Mas de Molino

2 HiPEACINFO  64

10

12

HiPEAC technologies pioneering medical breakthroughs

27 SME snapshot CVIS International: Delivering the compute behind scientific breakthroughs Gabriël Henriette 28 Innovation Europe Real-time healthcheck for coastal lagoons: SMARTLAGOON José M. Cecilia 30 Innovation Europe Smarter energy metering and social distancing precautions, thanks to SMART4ALL Katerina Labrakopoulou 32 Innovation Europe HealthyCloud: Defining the strategic agenda for the European Health Research and Innovation Cloud Esther Dorado Ladera 33 Technology opinion Start researching quantum logic and algorithms now Koen Bertels, Aritra Sarkar and Imran Ashraf 35 Technology opinion Architecture of large-scale AI systems Kemal Delic and Martin Walker 37 HiPEAC futures Electrifying futures: The DATE Young People Program HiPEAC internships: Optimizing vision-based embedded devices at Sundance


welcome

19

Silexit strategy: From spinoff to acquisition

24

Speeding up drug discovery simulations with GPUs

28

Real-time healthcheck for the Mar Menor

I am writing this editorial message on my way back from our annual ACACES summer school. Back in April 2021, we decided to organize ACACES as an in-person event. Several people contacted me to express concerns that it was too early, but we decided to go for it. We asked all participants to be fully vaccinated to reduce the risk and have more freedom. Now that the event is over, we know that we made the right decision, as the experience exceeded everybody’s expectations. For almost all of the 159 attendees, it was their first international in-person event in a year and a half. It felt great to interact with each other until late in the night, to have meals with participants from all over Europe, and to hang out in the charming town of Fiuggi while eating a gelato. There could be no bigger contrast with last year. Does this mean that everything was back to the old normal? No, it wasn’t. We had to distance, we had to wear masks while moving inside buildings and during the classes, and we had to regularly sanitize our hands. But apart from that, it was a summer school like those in the good old days: courses from top experts, long discussions with the participants, fabulous Italian food and without being constantly reminded that the virus HiPEAC is the European network on high performance embedded architecture and compilation.

was still among us. I am glad that we had the courage to organize this in-person event to serve the community, and that we learned a lot about organizing COVID-safe events at the same time. It is clear for HiPEAC: all future international networking events will be in-person. We had enough virtual networking. COVID is still here, but we should not let it control our lives, which means that we have to learn how to live with it. Vaccination in combination

hipeac.net @hipeac

with responsible behaviour is the cheapest and most effective way to live with the virus. hipeac.net/linkedin

HiPEAC has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 871174. Cover image: © Aoleshko | Dreamstime.com Design: www.magelaan.be Editor: Madeleine Gray Email: communication@hipeac.net

In the meantime, with healthcare still high on the agenda, we’ve brought together a range of stories in this issue showing how the advanced computing solutions developed by the HiPEAC community are promoting health and wellbeing. From wearables to support social distancing to the supercomputers driving the discovery of new medicines, it is no surprise to see HiPEAC technologies at the forefront of biomedical research and innovation. I hope you enjoy reading it, and look forward to seeing you at one of our future in-person events Koen De Bosschere, HiPEAC coordinator HiPEACINFO  64 3


HiPEAC news

Photo credit: © Tristan Deschamps

Bienvenue à Lyon – Welcome to Lyon! Situated at the confluence of the rivers Saône and Rhône, Lyon is France’s third largest city. Famous for its cuisine, with one of the highest concentrations of restaurants per capita in France, Lyon also has a proud history of technological innovation, such as the invention of the cinematograph by the Lumière brothers. We caught up with Computing Systems Week host Ian O’Connor, director of the System-on-chip, Systèmes embarqués et Objects Connectés Research Group (GDR-SOC2) at the Centre national de la recherche scientifique (CNRS), to find out more.

Tell us about the computing ecosystem in

image processing and parallel computing, to robotics, networks,

the Lyon area.

embedded and distributed systems, emerging technologies and computing paradigms.

Lyon is home to MILYON, a centre of excellence in the fields of mathematics and

Many industrial sectors are represented in Lyon, in such fields as mechanical engineering, textile, health, chemical engineering and pharmaceuticals, and make the city an ideal testbed for wide-

and eco-technologies focusing on multidisciplinary and multi-

scale deployment of the internet of things (IoT) and computing

scale research integrating physics, chemistry and engineering.

systems, as exemplified by the fast-growing and dynamic TUBÀ

Together, these centres of excellence bring together well over

living lab dedicated to the development of urban projects relying

1,000 researchers from over 25 research units located in Lyon

on digital technology and data science, as well as initiatives at

and Saint-Étienne. The topics of research are wide ranging –

the utilities company EDF / Smart Electric Lyon, Orange Labs,

from data and knowledge, algorithms and computer arithmetic,

Cosmo Tech and so on.

Place Bellecour and the iconic basilica on Fourvière hill

A trip to Lyon’s old town is a must

Photo credit: © Franck Chapolard

Photo credit: © Tristan Deschamps

theoretical computer science as well as iMUST, a centre of excellence in the fields of materials, processes

4 HiPEACINFO  64


HiPEAC news We are also just over an hour's drive away from Grenoble which

What should CSW participants see and do while they’re in Lyon?

enables us to collaborate closely with the major semiconductor manufacturing facilities, fabless design companies and computing

Admire the beautiful Renaissance architecture in Vieux Lyon,

research labs which are located there.

where semi-hidden passageways called ‘traboules’ connect courtyards with the Saône. Taste the food and wine from the city,

What is GDR-SOC2? What are your main areas of focus?

France’s culinary capital and home to the late French chef Paul Bocuse. Walk up Fourvière hill to the iconic, ever-visible basilica

GDR-SOC2 is a state-funded (CNRS) national research network

to the west of the city and ascend the north tower for one of

(Groupement de Recherche) focusing on systems-on-chip,

Lyon’s best photo ops. Let the millennia of history hit you as you

embedded systems and the IoT. It brings together around 800

sit in the Roman theatre of Fourvière and visit the Gallo-Roman

researchers and academic staff from over 65 national laboratories

museum located nearby.

working on: • embedded high-performance computing • analogue, mixed-signal and radio-frequency (RF) design • system security and integrity • the IoT • emerging paradigms and technologies • methods and tools • embedded artificial intelligence How do you think GDR-SOC2 can best collaborate with HiPEAC? Our goal is to bring together researchers to study and propose new approaches for the design and validation of embedded systems for IoT, edge computing and embedded artificial intelligence. The current challenges are numerous: reducing energy consumption in embedded systems to extend battery lifetime and in exascale computing to reduce carbon footprint, guaranteeing the safety and integrity of electronic systems, reducing the costs of designing and validating embedded computing systems and ensuring the suitability of integrated circuits in IoT applications, for example. We think that, between GDR-SOC2 and HiPEAC, we are facing many common challenges such that organizing and participating in events such as Computing Systems Week gives us the opportunity

19th-century architecture (including the opera house, city hall and refurbished Hotel Dieu), wide plazas, shops, cultural institutions, restaurants, cafes, bars and nightclubs. And relax in Parc de la Tête d’Or north of the centre and one of the largest urban parks in France, complete with a zoo and botanical garden within its boundaries.

FURTHER INFORMATION:

GDR-SOC2

gdr-soc.cnrs.fr

Manuel de conversation Do you know your logiciel libre from your transformation numérique? Here’s our handy guide of key phrases for your stay in Lyon, inspired by a few famous French thinkers. English

French

Hell is other people’s poorly documented code

L’enfer, c’est le code mal documenté des autres

I am optimizing for the best in the best of all possible worlds

J'optimise au mieux dans le meilleur des mondes possibles

Photo credit: © Tristan Deschamps

Software is born free but every­ Le logiciel est né libre et partout where is in proprietary chains il est dans les fers propriétaires

Photo credit: © Tristan Deschamps

to foster exchanges, share ideas and set up projects.

Soak up the atmosphere in the Presqu’Île, a district of glorious

Discover Lyon’s ‘secret passageways’, the traboules

Find out how Lyon earned its reputation as a culinary centre HiPEACINFO  64 5


HiPEAC news

Join HiPEAC at EFECS 2021 The European Forum for Electronic Components and Systems (EFECS) takes place as an online event on 23-25 November. This year’s strapline is ‘Our Digital Future, with a focus on delivering a green, competitive future along the electronic components and systems (ECS) value chain in Europe, managing the impact of electronics for a green transition to digitalization. HiPEAC will be participating in the event with a virtual booth. EFECS is organized by the industry associations AENEAS, Inside Industry Association (previously known as ARTEMIS-IA) and EPoSS in association with ECSEL Joint Undertaking, the European Commission and EUREKA. The fourth edition of the ECS Strategic Research and Innovation Agenda (ECS-RIA), developed by AENEAS, Inside Industry Association and EPoSS, is now available for download. It covers micro- and nanoelectronics, smart systems integration and embedded systems of systems.

FURTHER INFORMATION:

EFECS 2021

efecs.eu

ACACES returns to Fiuggi September saw the celebration of ACACES,

ACACES participants have learned to

Meanwhile, the poster session gave students

HiPEAC’s annual summer school on advanced

expect top-quality technical courses from

a much-missed opportunity to showcase

computer architecture and compilation for high-

internationally

professors.

their work and get feedback from senior

performance embedded systems, in the Italian

This year was no exception, with open

participants, and the HiPEAC Jobs wall gave

spa town of Fiuggi. Bringing together over 150

hardware, compilation, real-time systems

attendees an overview of the most interesting

attendees, the event was a welcome return to

and hardware security among the topics.

jobs in advanced computing. Needless to say,

in-person networking events after a prolonged

The technical sessions were complemented

local regulations were observed to try and keep

hiatus due to the COVID-19 pandemic.

by an entrepreneurial track managed by the

everyone safe from COVID-19, with attendees

TETRAMAX Innovation Action, including

either being vaccinated or testing regularly,

“ACACES was one of the most engaging and

courses on entrepreneurial finance and

masks being worn by attendees, greater

stimulating events I have had the privilege to

intellectual property.

distancing and rigorous hygiene measures.

reconfigurable programming with FPGAs, how

Keynote talks by Irfan Mirza (Microsoft) and

“I met really nice people, I learned a lot

to exploit task-based parallel programming

Giuseppe Gentile (IngeniArs) gave insights

of new things and I presented my work to

models, and gained valuable insights on

into resilience and career development,

others. It was a blast!” Teodora Vasilas, PhD

power consumption” Adriano Pimpini, PhD

while

candidate at Lucian Blaga University of Sibiu

student, Sapienza University of Rome

researchers to discuss computing trends in

recognized

experience. I learned about the RISC-V ISA,

a

dedicated

session

allowed

the HiPEAC Vision. FURTHER INFORMATION:

hipeac.net/events/#/acaces

6 HiPEACINFO  64


HiPEAC news

HiPEAC partner Per Stenström awarded ACM SIGARCH Alan D. Berenbaum award

FPL 2021 Community Award to Michael Hübner and students In

September,

(formerly

Muhammed

Ruhr-Universität

Al

Kadi

Bochum),

Hector Muñoz Hernandez and Michael Hübner (both Brandenburg University of Technology Cottbus-Senftenberg) received the Field-Programmable Logic (FPL) 2021 Community Award for their paper ‘FPGU: A SIMT-Architecture for FPGAs’.

During the International Symposium on Computer Architecture (ISCA) Conference

Rosa M. Badia wins HPDC Achievement Award 2021, elected to ACM Europe Council

The paper outlines a graphics processing unit (GPU) architecture based on a field-programmable gate array (FPGA). Muhammed developed the open-source core including the simulation and compiler infrastructure, which has since been used in research work all over the world, including at companies such as Intel and

in June, Per Stenström was presented with the Alan D. Berenbaum Award for

In June, HiPEAC member Rosa M Badia

Distinguished Service, which is presented

(Barcelona

annually. Per was selected 'for the creation

became the first researcher working in

and successful implementation of the

Europe to receive the International ACM

TACO/HiPEAC journal-first review model,

Symposium on High-Performance Parallel

and for decades long exemplary service

and

contributions to the field'.

Achievement Award, which recognizes

Supercomputing

Distributed

long-lasting,

Computing

influential

Center)

(HPDC)

contributions

One of the original founders of HiPEAC,

to the field. The award was presented in

Per is a professor and head of the

recognition of Rosa’s innovations in parallel

Computer Engineering Division within

task-based programming models, workflow

the Department of Computer Science

applications and systems, as well as for

and Engineering at Chalmers University

her leadership in the high performance

of Technology in Gothenburg, Sweden. In

computing research community.

ThinkSilicon. Subsequent work by Hector has seen the development of a system-onchip for space applications, combining the RISC-V and the FPGA. Congratulations to everyone involved in this work! Project github:

github.com/malkadi/FGPU

addition to his internationally recognized career in computer architecture research,

Meanwhile, in July it was announced that

Per has launched startups including

Rosa been elected to the Association

ZeroPoint Technologies, based on an

for Computing Machinery (ACM) Europe

innovative technique for data compression

Council, allowing her to amplify the

in memory.

presence of European members and increase the visibility of women within the

On behalf of the HiPEAC community,

association.

congratulations! Congratulations on behalf of HiPEAC! Check out our video interview with Per on

innovation

and

high-performance

computing on #HiPEACTV: bit.ly/PerStenström_HPC_innovation

FURTHER INFORMATION:

https://bit.ly/Rosa_Badia_HPDC_award

HiPEACINFO  64 7


HiPEAC news

On the edge of trust With edge computing continuing to be put

hardware, software and communications, to

in place across a range of creative new

applications, methodologies and assurance.

applications, we caught up with ACACES 2021 tutor Martin Törngren to hear about

Why is trust so important in edge computing? How can the

TECoSA, a Swedish research centre focusing

community get involved in your activities?

on trustworthiness in edge computing.

While many envisioned applications in edge computing are relevant in the context of real-time analytics and feedback, most of the research

What is TECoSA? Why was it established?

today approaches edge computing from a ‘best effort’ perspective,

Edge computing, also referred to as ‘fog’, ‘distributed cloud’ and

focusing on expected / average system behaviour. If edge computing

‘cloudlets’, is computing related to the emerging needs and

is to be used in industrial applications, trustworthiness must be

opportunities for computations local to the applications. It draws upon

considered as a key design objective. Depending on the application,

advances in cloud computing, communications, artificial intelligence

systems must go beyond the provisioning of average system

(AI) and human-machine interaction, such as augmented reality (AR).

behaviour: as well as considering predictability, latency, availability

Thanks to the fundamental improvements edge computing offers in

and reliability, they should take safety and security into account.

latency, bandwidth and energy efficiency with respect to offloading application workloads, it has become a topic of significant interest

In December, TECoSA will be hosting SEC 2021: Trustworthy Edge

both in academic and industrial research.

Computing Workshop as part of the ACM/IEEE Symposium on Edge Computing in San José, California, United States. The workshop will

In response to the growing importance of edge computing, we

focus on trustworthy edge computing systems and applications,

inaugurated the TECoSA (Trustworthy Edge Computing Systems and

including predictability in computation and communication, models

Applications) research centre in March 2020. TECoSA deals with

and algorithms for safety assurance, security and privacy for systems

challenges of cross-domain relevance, with a particular focus on

and applications, federated learning, and resource allocation.

safety, security and predictability challenges. It is hosted by KTH Royal Institute of Technology and collaborates with a number of industrial partners. With base funding from Vinnova, the Swedish innovation agency, as well as from industry and KTH, the long-term future of TECoSA is assured. What are the main technology areas covered by TECoSA? Within trustworthy edge computing, TECoSA is domain- and technology-agnostic. In the testbed currently being developed at the KTH campus, we evaluate all kinds of technologies, distributed computing concepts and architectures. We also aim to boost innovation in edge computing systems and applications, all the way from

8 HiPEACINFO  64

FURTHER INFORMATION:

SEC 2021: Trustworthy Edge Computing Workshop bit.ly/SEC21_workshop Subscribe to the TECoSA a newsletter via the centre’s website: tecosa.center.kth.se TECoSA seminar series: tecosa.center.kth.se/tecosa-seminar-series


HiPEAC news

Book: Security of Ubiquitous Computing Systems

Dates for your diary

Editors: Gildas Avoine and Julio Hernandez-Castro Nicolas Sklavos, University of Patras, Hellas This book presents knowledge acquired as part of the European Union COST Action CRYPTACUS (Cryptanalysis in Ubiquitous Computing Systems), which aimed to improve and adapt existent cryptanalysis methodologies and tools to the ubiquitous computing framework.

HiPEAC 2022 17-19 January 2022, Budapest, Hungary

An open-access publication, the book sets out to give readers an

Calls for workshop papers currently ongoing

understanding of current challenges and solutions in security for

Sponsorship and exhibition opportunities available

ubiquitous computing systems. Here, ‘ubiquitous computing systems’

hipeac.net/2022/budapest

is used to refer to small, embedded devices with serious constraints in terms of memory and processing power, typically no batteries but good connection capabilities, and, frequently, a number of sensors. This definition is, of course, flexible. Electronic passports, contactless transportation cards, personal assistants, as well as connected cars and fridges all fall within this definition. The cryptanalysis implemented lies along four axes: cryptographic models, cryptanalysis of building blocks, hardware and software security engineering, and security assessment of real-world systems. The authors are world-class researchers in security and cryptography.

EFECS 2021: European Forum for Electronic Components and Systems 23-25 November 2021, virtual efecs.eu ASPLOS 2022: Architectural Support for Programming Languages and Operating Systems 28 February - 4 March 2022, Lausanne, Switzerland HiPEAC Paper Award conference asplos-conference.org DATE 2022: Design, Automation and Test in Europe 14-23 March 2022, Antwerp, Belgium and online Various calls for contributions ongoing date-conference.com FCCM 2022: IEEE Symposium on Field-Programmable Custom Computing Machines 15-18 May 2022, New York, NY, United States HiPEAC Paper Award conference Abstract submission: 3 January 2022 fccm.org ISCA 2022: The International Symposium on Computer Architecture 11-15 June 2022, New York, NY, United States HiPEAC Paper Award conference Abstract submissions: 16 November iscaconf.org/isca2022 PLDI 2022: ACM SIGPLAN Conference on Programming

FURTHER INFORMATION:

bit.ly/Ubiquitous_Security

Language Design and Implementation 22-24 June 2022, San Diego, CA, USA HiPEAC Paper Award conference Paper submissions: 19 November pldi22.sigplan.org HiPEACINFO  64 9


HiPEAC Voices Alfonso Valencia (Barcelona Supercomputing Center) is an internationally renowned researcher in the field of bioinformatics. We caught up with him to find out about personalized medicine, digital twins in biology, how artificial intelligence is revolutionizing biomedicine and more.

“For the first time, we have real biological data with which we can create digital twins” What exactly does personalized medicine

one type of cancer – and therefore the

involve?

drugs used to treat it effectively – may be the same as those in another, as was the

A topic which has been spoken about for

case in a study where breast cancer was

some years now, personalized medicine

treated with drugs originally intended for

basically

pancreatic cancer.

encompasses

techniques

of

genomic, precision and data-oriented medicine. Thanks to advances in medical

By creating a highly personalized patient

instruments and devices, we can now

profile, it is possible to find the most

measure things that we were unable

suitable treatment for that patient, with

to measure in the past. This means that

applications beyond cancer. There are

medicine is increasingly being practised

hereditary diseases, for example, where

with

some patients experience very severe

more

information

and

more

precision, and thereby can become more

symptoms

while

others

have

minor

and more tailored to people’s individual

symptoms. Personalized medicine allows

characteristics.

us to better understand why this happens.

As an example, many drugs currently used

How is the concept of digital twins being

to treat disease have never been tested

applied in biomedicine?

on a specific demographic group, such

“Machine learning – from natural language processing to image recognition – and cryptography are two areas of vital importance in this field”

as older people. This could have serious

The concept of digital twins, which has

consequences, given that our metabolism

been around for a long time in engineering,

changes as we age.

is now being increasingly applied within the field of personalized medicine. In

Personalized medicine also provides a

engineering, digital twins are created using

more holistic way to investigate different

real data; today, for the first time, we have

therapies. For instance, the most common

real biological data with which we can

way of approaching cancer research

create something similar. It’s important

until now was to study each type of

to note, however, that biological systems

cancer individually. Researchers chose

are much more complex than mechanical

a particular type of cancer, identified a

systems, embodying many more elements;

drug, and undertook a large number of

the system of a human being is far more

clinical trials.

heterogeneous and less predictable than that of a satellite, for example.

With personalized medicine, however,

10 HiPEACINFO  64

researchers can probe how the cancer

We are still missing many parts of the

actually develops and focus on the

puzzle, but two technologies in particular

mutations which take place. This has led

have been fundamental in enabling this

to the discovery that the mutations in

development. First, the ability to sequence


HiPEAC Voices

individual cells, allowing us to view what

How is artificial intelligence revolutio­

that this achievement will have; it will

is going on in each cell. Second, the

nizing biomedical research?

undoubtedly

revolutionize

biology

research. This year, DeepMind published

synthesis of artificial organs, allowing us to try out different drugs and carry out

The roots of what we call ‘artificial

the algorithm – implemented on neural

experiments on organs. Of course, this

intelligence’ can be traced back to research

networks – with which they managed to

implies a serious computational challenge,

into natural language processing in the

solve this problem. An invention which

given the necessity of simulating billions

1960s and 1970s. Over half a century

required much imagination, the solution

of cells as they grow and differentiate.

later, machine-learning technologies are

is partly intuitive, rather than being based

delivering spectacular advances, partly

on brute force alone.

of

thanks to a change in approach. In the

Excellence (CoEs) are doing relevant work

area of natural language processing,

While artificial intelligence is therefore

in this area: BioExcel, which focuses on

for example, one of the main topics

driving incredible progress in biomedicine,

simulations at the biomolecular level;

investigated within my department, rather

bias is a major issue, and one which already

PerMedCoE, which undertakes simulations

than trying to get the system to ‘learn’ an

has a problematic history in biomedical

at the cellular level; and CompBioMed,

entire language, today’s systems simply

research. In the Life Sciences department

which is working on simulations of organs.

learn its structure.

at

Three

European-funded

Centres

BSC,

we

have

a

group

named

BioInfo4Women. As well as promoting What are the main requirements of

A high-profile example of this shift in

female

bioinformaticians,

this

group

bio­ medical applications in terms of

approach is AlphaFold, the programme

examines scientific problems related to

computing technology?

developed by DeepMind which in 2020

bias in biomedicine. One striking example

solved a scientific problem first posed

is the fact that proportionally more women

It goes without saying that serious

50 years ago. Specifically, this software

than men die of a heart attack, as women

computational horsepower is required

programme managed to predict, with

tend to exhibit different symptoms which

to carry out these kinds of analysis. In

very high precision, how proteins fold

are not as well known or understood,

general, the most important consideration

and form the three-dimensional structures

and they are at risk of not receiving the

for biomedical applications is the ability

which determine their function.

required attention or treatment.

to process vast amounts of data; related to this, security needs to be taken extremely

Predicting

strings

Similarly, there are many well-known

seriously, in order to prevent confidential

of amino acids will take is extremely

examples of bias in artificial intelligence

patient data becoming compromised.

difficult, given that exploring all the

systems. In the end, the quality of the

Hence machine learning, particularly

different structures possible would need

system and the trustworthiness of its

on deep neural networks – from natural

an impossible amount of computational

results depend on the quality of the data

language processing to image recognition

time. It is hard to overstate the importance

and algorithms used to build it.

what

form

these

– and cryptography are two areas of vital importance in this field. As for hardware, systems used for biomedical applications are generally federated, complex, closed and secure. They need large amounts of memory and rely on accelerators such as graphics processing units (GPUs) to carry out the necessary processes. Another key consideration is the ability to carry out the computation at the most appropriate point on the compute continuum; there may be instances where it is important to process the data locally, on small devices at the edge, while simulations will likely

Alfonso Valencia with members of the life sciences department at Barcelona

require a large computing infrastructure.

Supercomputing Center HiPEACINFO  64 11


Special feature: Healthcare and biomedicine

First aid

The expertise of the HiPEAC community spans the compute continuum from the nimblest devices at the edge to the mightiest supercomputers. With the COVID-19 pandemic focusing the world’s attention on health issues as never before, this article explores a few examples of computing innovations within the community, enabling everything from smart wearables to complex simulations.

GIVING SURGEONS A HAND IN THE OPERATING THEATRE There are situations where a standard human-computer

allowing surgeons to control data visualizations with pre-defined

interface, such as a touchscreen, keyboard and / or mouse,

hand gestures. ‘Motion tracking enables more precise virtual

just aren’t practical. Martin Žagar (RIT Croatia) briefs

movement, rotation, cropping, spatial locking and measuring, as

HiPEAC on a solution for surgeons his team delivered in

well as slicing through datasets,’ Martin says. ‘We developed a

collaboration with Bitmedix.

software framework for three-dimensional (3D) virtual navigation that can interface with cameras and provide a set of standardized

Picture the scene: you’re a consultant undertaking complex

methods for medical applications, such as hand gestures.’

surgery, and you need to get a better view of a segment of the patient’s tissue at a critical juncture in the operation. ‘Currently,

To provide the most immersive experience, the team used a

surgeons rely upon assistants to navigate the data for them,

camera for depth and motion tracking with active stereo depth

using a keyboard and mouse,’ explains Martin. ‘This involves

resolution, precise shutter sensors, and a range of up to two to

oral communication, which could result in misunderstandings,

three metres, giving the surgeon more freedom to move around.

the navigation might not be precise, the assistant might zoom in

‘We used a machine-learning pipeline consisting of multiple

too much or too little, and the whole process takes time – which

models: a palm-detection model that operates on the full image

might be of crucial importance during surgery.’

and returns an oriented hand bounding box, and a hand landmark that operates on the cropped image region defined by the palm

Take virtual endoscopy, for example. ‘In this kind of procedure,

detector and returns high-fidelity 3D hand keypoints,’ explains

a surgeon needs to be very precise and examine the data from

Martin. ‘Our approach uses a single deep neural network and

different angles and directions. This entails a lot of communication

can detect left and right hands of various sizes, with an average

with the assistants, and the virtualizations won’t be as precise as

precision of 95.7% in palm detection.’

they could be.’ After carrying out palm detection on the whole image, the hand To solve this problem, Martin’s research group at RIT Croatia teamed

landmark uses regression to perform precise keypoint localization

up with Croatian company Bitmedix to create a hardware sensor

of 21 3D coordinates inside the hand regions detected, according

device controller that supports hand and finger motions as input,

to Martin, as shown in figure 1.

12 HiPEACINFO  64

Photo credit © Aoleshko on Dreamstime.com

How HiPEAC technologies are powering medical breakthroughs


Special feature: Healthcare and biomedicine Figure 2 shows the functionality of the hand and gesture module API. ‘The system’s input is a 2D RBG image and 3D cloud (XYZ cloud points) provided by an Intel Realsense D415 camera,’ says Martin. ‘Data are processed by the hand tracking module, where the hand’s centre is generated in 2D image pixel data and 3D realworld data. Once the position of the hand is detected, the gesture recognition module estimates the gesture the hand is making.’ The information generated within the hand and gesture (H&G) module is stored in the H&G interface class, which represents a package that is sent to two different interactors, says Martin. ‘The first, the system mouse interactor, provides basic interaction with the mouse (mouse movement, grabbing, and selecting). The

Figure 2: Hand and gesture module API

second, the DICOM viewer interactor, provides “in the air” control of the DICOM viewer application (2D and 3D scene manipulation,

The resulting solution, which allows bone structures to be

changing DICOM parameters, 2D and 3D measuring, etc.).’

displayed in true 3D and bone/soft tissue structures to be viewed from all perspectives, includes a number of benefits, according to Martin. ‘For example, it makes it easier to precisely recognize pathologies, enables non-invasive procedures, helps determine the optimal point of surgical access, and improves the overall safety and efficiency of surgical interventions.’ This particular solution was developed within the EIT Health RIS Innovation programme, where RIT Croatia was an academic partner and project leader and Bitmedix was a project partner. ‘We are currently in the process of acquiring strategic partners for developing market solutions and commercialization as a next

Figure 1: The gesture ‘two’, demonstrating the 21 keypoints on the

step,’ adds Martin.

hand’s surface

ANTIRETROVIRAL DRUG CAN BE EFFECTIVE TREATMENT FOR CANCER Discovery enabled by HPC A group of researchers including HiPEAC member Horacio Pérez-Sánchez (Universidad Católica de Murcia-UCAM) has identified that a drug currently used to treat HIV may also be an effective treatment against certain types of colorectal cancer. In line with previous studies, the group found that a particular protein, Fascin1, was directly related to the invasiveness of tumour cells associated with worse survival rates in a number of carcinomas. After virtually screening a library of 9,591 compounds, they identified the anti-retroviral raltegravir as

‘We have been developing and improving computational drug

a potential Fascin1 blocker. Experimental characterization

discovery methods on high-performance computing (HPC)

subsequently demonstrated that the invasive capacity of

architectures for a number of years now,’ comments Horacio

colorectal tumour cells was notably impaired when the drug

Pérez-Sánchez. ‘These methods are applied to different

was present, without damaging healthy cells.

pharmacological contexts – in this case colorectal cancer. They have been continuously optimized for HPC; for instance, they now run over singularity containers.’ HiPEACINFO  64 13


Special feature: Healthcare and biomedicine Supercomputing infrastructure was provided by a number

also investigating whether this discovery has implications for

of HPC centres across Europe to undertake simulations

other forms of treatment. See paper in ‘Further reading’, below.

and screening of the pharmacological compounds. For the HPC simulations, the workload manager Slurm was used,

FURTHER INFORMATION:

while Autodock Vina along with other in-house scripts and

B. Alburquerque-González et al. ‘The FDA-Approved Antiviral

software along with singularity containers were used for the

Raltegravir Inhibits Fascin1-Dependent Invasion of Colorectal Tumor

computational drug discovery aspects.

Cells In Vitro and In Vivo’ Cancers 2021, 13(4), 861

A patent for the application of this antiviral compound in the colorectal cancer context has been filed and the group is now contacting companies with a view to licensing it. The group is

doi.org/10.3390/cancers13040861 Podcast episode detailing computation process for this research: bit.ly/CDD_blind_docking_podcast

As shown in these images, the stimulation of the actin-bundle protein Facsin 1 is prevented by raltegravir (taken from Figure 4 from doi.org/10.3390/cancers13040861)

COVID-19 VIRAL MODELLING AND INVESTIGATING CANCER DRUG COMBINATIONS WITH PERMEDCOE PerMedCoE, the HPC/Exascale Centre

can study perturbations on metabolism, signalling pathways

of Excellence in Personalised Medicine,

and cell populations, among other areas. We caught up with

aims to harness the power of exascale

Arnau Montagud and José Carbonell (Barcelona Super­

computing to solve problems in the field

computing Center) to find out more.

of personalized medicine. While exascale computing promises faster, more capable   computers, much of these machines’

Drug combinations for cancer treatment

potential will go unused unless current

effects, notably from the high concentrations of the drugs used,’

high-performance

says Arnau. ‘If better combinations of drugs were found, this could

computing

(HPC)

‘As is well known, cancer drugs often result in unpleasant side

software is adapted.

allow drug doses to be reduced and resistances to be avoided.’

To avoid this, PerMedCoE is testing four

To contribute to this field, PerMedCoE is studying the effects of

software tools for use on future exascale computers. These

different drug combinations and concentrations in the simulation

four tools will be applied in different use cases, reflecting

of cell-line specific models. ‘Rather than taking generic signalling

a broad range of computationally demanding real-life

models found in online resources such as the Atlas of Cancer

biomedical scenarios where cell-level models are used. Cell-

Signaling Network (ACSN), cell-line specific models capture the

level models study the mechanisms disrupted in cells where

specificities of a given cell line and how drugs interact with it,’

disease is present or when a drug is detected. These models

explains Arnau.

14 HiPEACINFO  64


Special feature: Healthcare and biomedicine

Simulation of mutants and heterogeneous cell populations. The PerMedCoE framework can simulate wild type cells (A) and study how knockouts, such as FADDs, alter the model’s behaviour (B). It also allows the study of different proportions of cells with an inhibited virus’ M protein, such as 50% of M knockout cells (C) and 95% of M knockout cells (D) Eventually, the aim is to integrate the results obtained by the

help explain why different patients experience varying levels of

BioExcel Centre of Excellence. With a focus on biomolecular

severity of the disease.’

simulations, BioExcel aims to identify and repurpose drugs for different diseases.

Scaling up software for cell-level simulations One of the key challenges within the project is to take existing

Multi-scale modelling of the COVID-19 virus and patient tissue

tools, developed in a range of programming languages, and

‘Thanks to tremendous, global efforts, we now have vaccines to

is our work on PhysiCell, which demonstrates the PerMedCoE

help prevent infection by COVID-19,’ says José. ‘However, once a

approach,’ says Arnau. ‘PhysiCell is a physics-based tool designed

patient is infected, we are still struggling to find the right drugs

to simulate the evolution and dynamics of a heterogeneous cell

for treatment.’

population – think of a tumour, for example. It uses an agent-

port them efficiently to an HPC environment. ‘A good example

based paradigm to model cell-specific behaviours, allowing us to Another PerMedCoE use case aims to help solve this by studying

simulate a wide variety of real-life scenarios.’

the COVID-19 disease using multi-scale models and single-cell data. ‘Multi-scale models consider different time and space

Written in C++, PhysiCell will need to be refactored to use the

scales. Our multi-scale model takes a lung epithelial cell layer

message passing interface (MPI) protocol so that the computations

– that is, the layer of cells lining the inner space of lungs –

and memory can be distributed over different machines. ‘Our

that responds to viral infection through signalling pathways.

strategy for PhysiCell focuses on dividing the physical space

The model allows us to organize the available knowledge

where cells are located in subdomains, each of which is managed

on molecular mechanisms, identify biomarkers and propose

by an individual MPI node. This allows us to scale up individual

therapeutic targets,’ explains José.

simulations significantly and achieve simulations of billions of cells,’ adds José.

In addition, the project aims to use single-cell data from patients to personalize these models. ‘Single-cell granularity allows

Find out more on the PerMedCoE website:

permedcoe.eu

researchers to study the diversity of different cells, their various maturation stages and how these affect their signalling and

PerMedCoE has received funding from the European Union’s Horizon 2020

behaviour,’ José notes. ‘By integrating these data, we hope to

research and innovation programme under grant agreement no. 951773. HiPEACINFO  64 15


Special feature: Healthcare and biomedicine

SPACE-BASED COVID-19 TELEMEDICINE SOLUTIONS Camilla Giunti, IngeniArs

the European Space Agency. This aims to integrate satellite technologies into E@syCare, a telemedicine system developed by IngeniArs which can autonomously acquire vital parameters and enable remote monitoring of patients, and which is certified as a class I medical device according to Directive 93/42/CEE. The new features – integration of a smartwatch with GPS and geo-tagging of measurements – provide healthcare professionals with advanced analysis tools to monitor the physical activity (steps, calories, heart rate) of their patients. They can also monitor the concentration of people infected with COVID-19 or who have certain symptoms in a given geographical area, promptly identifying any outbreaks. The system has been used in Tuscany by general practitioners, who have monitored more than 600 patients, reducing hospitalizations and deaths. In the

For over five years, Italian company IngeniArs has been

long term, it will allow the physical activity of chronic patients

working on innovative applications for telemedicine, focusing

to be tracked.

on the development, provision and integration of information and communication technology (ICT) solutions. IngeniArs offers advanced ICT solutions to remotely facilitate, organize and manage the personalized care of patients who are frail or have chronic conditions, by monitoring their clinical status and lifestyle. As a response to the COVID-19 emergency, IngeniArs is carrying out the SatNav E@syCare project in conjunction with

TIME TO THROW AWAY YOUR READING GLASSES? A prototype to correct presbyopia in real time

Presbyopia is not a new phenomenon but a natural condition that has always happened to humans as part of the ageing process. As

Juan Mompeán, Juan Luis Aragón and Pablo Artal, University

such, a number of commercial solutions exist to mitigate it: from

of Murcia

wearing reading glasses, the simplest and cheapest solution, to more advanced approaches such as surgery-implanted multifocal

Presbyopia – a word derived from the ancient Greek meaning ‘old

intraocular lenses. However, these solutions have significant

eye’ – refers to the progressive decrease in the accommodation

disadvantages. For example, none of them offers a continuous

range of the human eye until this capacity completely disappears.

focus range; instead, they are designed for only one, two or three

It is caused by the lens in the eye losing flexibility as we age.

focusing distances. Furthermore, when several focusing distances

While this process starts in the childhood, it is not normally

are combined into a single lens, the image contrast is reduced,

noticeable until people reach their 40s or 50s. At that age the

while images may also be deformed.

accommodative range falls below 3 dioptres and people start to have difficulty focusing on close objects. Nowadays, presbyopia

To overcome those limitations, we are developing a pair of

creates even greater problems than in the past, due to the longer

automatic glasses to dynamically correct presbyopia. To do this,

time we spend looking at objects in the near distance, especially

they incorporate two optoelectronic lenses, two cameras, the

electronic devices such as smartphones and tablets.

electronics to control the lenses and the cameras, a battery and a processing device. The glasses automatically correct presbyopia

16 HiPEACINFO  64


Special feature: Healthcare and biomedicine by tracking where the subject is looking in real time. Our approach uses a continuous focal range and is not limited to a fixed number of focusing distances. This is achieved by tracking both pupils and calculating the vergence – the movement of the pupils – of the subject’s eyes to precisely measure the looking distance. Using this distance, the computing device calculates the optical correction that needs to be applied by the optoelectronic lenses. Our current prototype uses a smartphone to perform all the processing of the images captured by the cameras in real time. A fast and accurate pupil-tracking algorithm is used to precisely calculate the current vergence and the correction required. The smartphone runs an OpenCL implementation of the binocular pupil tracking algorithm on its graphics processing unit (GPU),

Left to right: Pablo Artal, Juan Mompeán and Juan Luis Aragón, creators of the optoelectronic glasses prototype which can process around 25 frames per second, providing a very good experience for the subjects who have tried the prototype. The processing speed is crucial to ensure a smooth transition in the optical power applied by the optoelectronic lenses. Similarly, accuracy is important in order to avoid rapid changes in the lenses which would be uncomfortable for subjects. As a result, the visual performance of the device is excellent when compared with static solutions, such as reading The optoelectronic glasses prototype

glasses or progressive glasses. Given the wearable nature of the device, tight integration between the hardware and the software is of key importance in order to achieve the required performance at a reasonable power consumption. To solve that challenge, we are currently exploring different architectures to process the images and control all the hardware. Our options include the use of a field-programmable gate array (FPGA) to implement a specialized and energyefficient pupil tracking algorithm, which would be capable of sending the images wirelessly to an external computing device, or building a custom system with mobile chips.

“Our approach uses a continuous focal range and is not limited to a fixed number of focusing distances.”

The prototype uses a pupil-tracking algorithm to calculate the current vergence HiPEACINFO  64 17


Special feature: Healthcare and biomedicine

THE SENSE OF EMOTIONS: TULLY, A WEARABLE FOR HYPERACTIVITY DISORDERS A chance encounter led Marius and Adina Rus, the founders of StressLess, to develop a solution for children with hyperactivity disorders. ‘Our device for hyperactivity disorders, Tully, came about from our real-life experience,’ explains Marius. ‘My wife Adina and I moved to San Francisco in 2011 so I could study for an MBA at the University of California, Berkeley. While we were there, my wife – an economics professor at Babes-Bolyai University – took time out to volunteer at our daughter’s school. By chance, she started working with a child who had severe hyperactivity, and found it extremely rewarding. The principal enrolled her in a course for counsellors, and in no time she was volunteering at all three schools in the Albany district.’ After their return to Romania, Adina decided to study for a master’s in psychology while Marius worked on finance and strategy with startups. ‘One day, while she was working on her dissertation, we realized that the classroom management activity of a specialized counsellor could probably be done by an electronic device, which could support children around the clock.’ The idea for Tully was born. ‘Tully is a wearable device that serves as an early warning system, alerting the child when a high-intensity emotional event – or “meltdown” – is imminent, and guiding them through the calming exercises learned in therapy,’ explains Marius. ‘It works by continuously monitoring a series of biomarkers, combining them into a proprietary “agitation score”, and interpreting that score to predict emotional flares. Significant added value comes from the data, which is processed and available, through parents, to the child’s therapists, opening up new therapy possibilities and allowing them to evaluate the long term success of the therapy,’ he adds.

Developing the device brought its own challenges, however: ‘As we needed specific sensor inputs, we were unable to use offthe-shelf wearable devices and had to develop our own system. This was when we found out that there is a lot of truth in the saying “hardware is hard”,’ notes Marius. ‘Any iteration, debug or modification took longer and incurred higher costs than we’d been expecting. Being bootstrapped for most of the last few years, it took us over a year to get to the first functional prototype and start collecting data and working on the algorithms.’ Four years after the couple started working on the project, Tully is now nearing its launch: ‘We did use a lot of user input, and went through several versions, with different functionalities until we reached this one, providing the biggest value for all stakeholders (children, parents and therapists),’ says Marius. In 2020, the Tully technology was tested on a large scale for the first time, Marius says, including both observation and data collection in the psychologist’s office and unsupervised use by children during school hours. ‘We are continuing this testing this year, with the dual aim of confirming the utility of using Tully and improving the detection algorithm by adding artificial intelligence (AI) capabilities,’ he adds. As for the future, a crowdfunding campaign is currently underway to launch Tully as a consumer electronics device, with the aim of going live by March 2022, starting clinical trials and obtaining medical device certification afterwards. ‘Tully isn’t limited to hyperactivity disorders. There are a wide range of conditions where people could benefit from help in emotion recognition and management, from specific conditions like oppositional defiance disorder or autism, to general ones like depression and anxiety,’ explains Marius. ‘We are excited about the next stages of getting Tully onto the market and providing a wearable that could significantly help users.’ FURTHER INFORMATION:

The team that developed Tully 18 HiPEACINFO  64

mytully.com


Tech transfer

Founded by Maximilian Odendahl, Johannes Emigholz, Weihua Sheng, Jeronimo Castrillon and HiPEAC steering committee member Rainer Leupers (RWTH Aachen), Silexica aimed to take the pain out of heterogeneous system-on-chip programming through automation. Thanks to the hard work and persistence of the team, led by Max, its flagship suite of tools, SLX, was launched in 2014, and the company has been going from strength to strength. Recently, Silexica was bought out by Xilinx, the inventor of the field-programmable gate array (FPGA), programmable systems-on-chip (SoCs), and now the Adaptive Compute Acceleration Platform (ACAP). We caught up with Rainer and Max, the latter now at Xilinx, to find out more about this incredible success story.

Silexit strategy

How Silexica went from academic spin-off to acquisition by Xilinx Congratulations on the acquisition!

Max: We then successfully applied for an EXIST Transfer of

Let’s go back to the beginning.

Research grant, provided by the German federal government,

How did Silexica come about in

which supports outstanding research-oriented projects that

the first place?

involve expensive and high-risk resource development. This

Rainer: We started this as an academic research project around

allowed us to spin off the company, transfer the intellectual

2007. It was clear that multicore architectures would be around the

property (IP) and hire an initial team. So we didn’t start from

corner in virtually all embedded information and communication technology (ICT) domains, yet they lacked proper softwaredevelopment tool support. We had a lot of compiler technology and processor design knowhow from earlier projects and decided to take on the multicore programming challenge. After a series of fundamental PhD and master’s theses, the technology became mature enough to transfer, and even sell, to early adopters from industry. Another important milestone was a hands-on workshop for potential customers in Aachen, which provided us with important feedback and gave us the confidence that we really were on the right track. At the same time, we had an ambitious and diverse founding team in place, with Max in the cockpit, which is equally important.

Silexica founders (left to right) Rainer Leupers, Max Odendahl, Weihua Sheng, Johannes Emigholz and Jeronimo Castrillon HiPEACINFO  64 19


Tech transfer zero, and this enabled us to secure funding to continue building the team quite soon after the foundation. What made the product offer stand out? Rainer: I have seen quite often that programming tools for new hardware architectures are just an afterthought. The semi­ conductor market history is full of examples of powerful, yet ‘unprogrammable’, hence unusable, chips without business success. Our mission was to enable software developers to really take full advantage of the forthcoming heterogeneous multicores. We were able to demonstrate that this worked for complex real-life products, such as the Samsung Galaxy S6, which was among our first target devices. Another important use case was the ability to perform hardware/software co-optimization based on virtual prototypes, long before the actual hardware would be taped out.

doors of various administration departments and getting all their OKs. Universities aiming at boosting their technology transfer

Max: We were – and still are – the only ones not just using static

rate should definitely put more trust in the founding teams and

code analysis, but also adding dynamic source code analysis of

eliminate bureaucratic strings.

an application. This means we actually execute the application, enabling much deeper analysis capabilities. More data leads to

Max: I think all the usual pitfalls of a technical university spin-

improved application speedup and better actionable insights,

off applied to us. Talking way too much about technical features

with even more automation opportunities in the future.

instead of customer benefits, understanding product-market fit but not product go-to-market fit, hiring lots of engineers without

How did you manage the process of setting up the company?

proper product management support, scaling sales too early; the

What lessons did you learn in the process?

list continues.

Rainer: Fortunately, RWTH Aachen is a very founder-friendly university. Our institutes and upcoming startup teams enjoy a lot

If I had to pick one, I think a true lesson was getting a real

of entrepreneurial freedom. A key aspect is always to get the IP

understanding of the holistic approach of marketing and what

transfer right in legal terms, because investors will be sensitive to

it includes: not just corporate marketing, but also product

that. I had some experience from a previous spin-off (LISATek),

planning, product management and product marketing. So I

so I knew the major steps, and altogether the process was very

would definitely recommend talking to lots of people, listening

smooth in my view.

carefully and getting some experienced startup founders or executives as mentors who’ve been on this journey already and

Nowadays it’s probably even easier thanks to the recent

understand the challenges associated with building something

foundation of RWTH Innovation GmbH. This provides a one-

from the ground up.

stop shop for founders and avoids the need for knocking on the What was the most challenging aspect about setting up the company? Were there times when you felt like giving up? Max: I think there are challenges every single week, from negative feedback from a customer, to the key hire you didn’t make, a last-minute budget cut from your largest customer or a worldwide pandemic. But there are lots of rewarding moments every week as well, such as securing a big deal, hiring someone who will bring you to the next level and seeing your strategy actually working. Ultimately, you need the conviction that your product or solution is needed and you are on the right path. And even though we had various pivots throughput the years, giving up was never an option. RWTH Aachen provides a founder-friendly culture 20 HiPEACINFO  64


Tech transfer Do you have any tips on how to scale up a company?

Max: I think the various options were clear from day one, and

Rainer: That’s a tough question. Is it better to have a big piece

we didn’t exclude anything there, from being acquired by a

of a small cake or vice versa? I have seen various companies that

semiconductor company, becoming a new electronic design

bootstrapped slowly, without major investment rounds, over a

automation (EDA) player or doing an IPO. We indeed did have

long period and that eventually had a very profitable exit. On the

interesting discussions around the creation of a special-purpose

other hand, if your initial ambition is a unicorn, then fast growth

acquisition company (SPAC) – a shell company created to acquire

is a must. In deep tech particularly you need international

a private company – at the same time we were having merger and

presence, and that in turn implies scaling up as soon as possible.

acquisition (M&A) discussions. But Rainer is absolutely right: the

In the end, it’s a matter of the team’s attitudes and opportunities

key focus should be on adding customer value and showing why

popping up along the way.

you exist; everything else will slot into place.

Max: I think it is key to sit down right from the start with your

What are you most excited about for the future? How do you

team and discuss what type of company you want to be. Why

think the Xilinx buyout will help Silexica develop in the future?

you exist in the first place, how you fit into your customers’

Rainer: Some people remarked that they don’t like to see

value chain, what niche you can win first, how to expand, and

European Union (EU) startups often being acquired by companies

what the overall ambition and vision of the company is. Then go

from the United States. In my view, they often overlook we are

backwards to plan how to get there: what milestones do we need

living in a very globalized world. It’s also not unusual for EU

to hit, what kind of team do we need, when do we need money

companies to perform overseas acquisitions as well. At the end

and how much do we need to execute that plan.

of the day, what counts is where new workplaces are created and where taxes are paid to serve our economy. In that sense,

Ensuring that your product strategy and cooperate strategy are

the acquisition is a great move, hopefully inspiring many future

always well aligned and feed off each other – such as hitting

HiPEAC founding teams.

important product or customer milestones – is also important in order to raise the next round of funding (assuming you are on a

Max: I think this acquisition is a perfect example of this kind

growth track backed by venture capital).

of boost to the local economy. Software programmability is imperative to Xilinx’s long-term goal to accelerate the path

What was the company’s exit strategy? Did you ever have any

from software to application-optimized hardware systems. The

plans to go public?

Silexica technology complements the existing solution and

Rainer: Initial public offerings (IPOs) in deep tech are risky, costly

roadmap and will accelerate their roadmap to attract software

and tedious, thus pretty rare these days. So the basic options left

developers.

are to establish a self-sustainable new long-term business or to target an attractive acquisition by a larger corporation. Venture

Starting with our existing stellar team, we are now planning

capital firms, with their fixed-term investment funds horizon,

to grow the Cologne office significantly in the future and build

clearly have a preference for the latter, and this is how Silexica

a centre of excellence for compilers, machine learning and

exited. My advice to founders would be to focus on the business

artificial intelligence more generally. Anyone with an interest and

development, while ensuring maximum visibility and just enjoying

expertise in those areas, please check out our open positions or

the day-to-day fun in ramping up the company. Serious exit

reach out to me.

considerations will follow automatically when the time is right.

The Silexica team in 2019 HiPEACINFO  64 21


EU services to support your innovation pathways Does your European Union (EU) project need an exploitation plan? Are you interested in commercializing your research, but not sure where to start? Do you find navigating intellectual property rights and licensing agreements daunting? Many universities and research centres now benefit from support structures such as innovation management or technology transfer offices. Even so, tracing a path through all the innovation steps can seem overwhelming for researchers trying to bring their research to market. For this reason, the EU has provided a set of free resources and support programmes to help you get going.

IP Helpdesk iprhelpdesk.eu The Intellectual Property (IP) Helpdesk provides resources, guidelines and training material to help organizations to manage their IP effectively, helping you promote your businesses or maximize the impact of research and innovation projects. Guides are available on the following topics, among others: IP commercialization IP and contracts IP management in international business IP in H2020 IP in Europe

Research impact is becoming increasingly important in the Horizon Europe programme, with the European Commission keen to see the results of its publicly funded research becoming marketable goods and services that contribute to Europe’s prosperity.

However,

commercializing

new

techno­ logies

requires a whole new skillset and knowledge base compared to researchers’ day jobs of carrying out basic research. Luckily, the commission has created a suite of tools and support

You’ll also find factsheets if you need to check common issues relating to IP rights, such as: IP enforcement, trade secrets, commercializing IP and licence agreements, non-disclosure agreements as a business tool, trademarks, copyright and related rights in the Digital Single Market, among many others. There is also a comprehensive

glossary with definitions of

common IP terms, while training courses and webinars related to IP are available via the

events section in the website.

services to help your technology on its journey from the lab to the real world. We’ve rounded up a few of them to help you along.

Horizon IP Scan bit.ly/EU_Horizon_IP_Scan Horizon IP Scan is a tailored, free-of-charge support service designed to help European start-ups and other small / medium enterprises (SMEs) involved in EU-funded projects to efficiently manage and valorize IP in collaborative research and innovation (R+I) efforts.

22 HiPEACINFO  64

Photo credit: © Cammeraydave on Dreamstime.com

HiPEAC innovation booster


HiPEAC innovation booster

The Horizon IP Scan team draws upon a network of IP experts from across Europe to perform an individual assessment of SMEs’ intangible assets. The service aims to: • Show how existing IP can be protected when starting on a research and innovation project with multiple partners. • Assist SMEs in developing a shared strategy with project partners to manage and exploit new IP generated jointly as part of their projects.

Horizon Results Booster horizonresultsbooster.eu Effectively disseminating and exploiting your research results means that publicly funded European research can have real impact on society. But how do you go about it? The Horizon

business innovation consultancy and help forming international

Results booster provides a package of personalized services to

partnerships.

amplify the added value of EU Framework Programmes. Head here if you’re working on a Framework Programme research project and in need of:

Horizon Results Platform

• Help creating a portfolio of project results, a dissemination plan and

bit.ly/EU_Horizon_Results_Platform

an exploitation strategy. • Assistance in creating a business plan to bring your research results

According to the European Commission, results of research and

closer to the market. This encompasses market analysis, competitor

innovation projects include data, knowledge and information,

identification, business strategy, operations plan, action plan and

along with any related rights, such as intellectual property

estimation of time to market, for example.

rights. The commission has created a new category, named ‘key

• Support in ‘go-to market’ activities such as pitching, IPR support,

exploitable results’, for results that have high potential for being

innovation management, commercial development, as well as

taken up, used and generate value, whether this be contributing

support in the creation of spinoffs and startups.

to the downstream value chain of a product, or providing input to policy or education. To highlight these assets, the commission has created the Horizon

European Enterprise Network (EEN) een.ec.europa.eu

Results Platform, allowing project participants to upload key exploitable results from the participant portal.

You’ve managed to launch your spinoff or startup, but how do

Have you managed to get your research results to market? The HiPEAC

you scale up? Get answers and support from the Enterprise

Technology Transfer Awards are open now – apply by 22 November.

Europe Network, the largest network of its kind in the world,

Further information:

hipeac.net/tech-transfer

which helps businesses continue to innovate and grow on an international scale. Comprising over 3,000 experts from more than 600 organizations, the network is active in more than 60 countries. Its members include technology poles, organizations supporting innovation, universities and research institutes, regional development organizations, and chambers of commerce / industry.

`

Together, these provide services for individual business, including advice and support on how to grow internationally,

HiPEACINFO  64 23


The COVID-19 pandemic has generated a heroic response from scientists across the world. In this article, Leonardo Solis-Vasquez (TU Darmstadt), Andreas F. Tillack (Scripps Research), Diogo Santos-Martins (Scripps Research), Andreas Koch (TU Darmstadt) and Stefano Forli (Scripps Research) describe how they helped improve the processing speed of drug discovery simulations by 50 times, and how their optimized codes have been set to work on computers all over the globe, from micro boards to mega machines.

Speeding up simulations for drug discovery with Autodock-GPU has

accelerate these long-running simulations

threaded AutoDock4 baseline (R9-290X

demonstrated the need for ways to

using graphics processing units (GPUs),

GPU vs. i5 CPU).

quickly discover new drugs. One of

leading to the release of AutoDock-GPU,

these approaches is molecular docking

which is the subject of this article.

The

advent

of

SARS-CoV-2

molecules that, when attached to a

The first AutoDock-GPU implementation

AutoDock-GPU vs. COVID-19, from supercomputers to Raspberry Pis

viral protein, prevent the latter from

resulted from a joint effort between Scripps

The outbreak of the COVID-19 pandemic

infecting healthy cells. Molecular docking

Research and a team at the Technical

prompted researchers from the Oak Ridge

interactions are modelled by atomic forces

University of Darmstadt. AutoDock-GPU

National Laboratory in the United States

being applied to molecular shapes. The

was specifically developed for leveraging

to use AutoDock-GPU on the Summit

complexity of the necessary calculations,

both multicore central processing units

supercomputer, currently the second fastest

and the large number of molecules to

(CPUs) and highly-threaded accelerator

computer worldwide according to the

be assessed, require high-performance

devices such as GPUs. This original

TOP500 list. This effort involved NVIDIA

computing. A larger number of computed

version was developed in OpenCL (Open

and Scripps Research in a collaboration

results can be exploited afterwards to

Computing Language), which, following

to port AutoDock-GPU to the CUDA-

narrow the search space for the costly and

the philosophy of ‘write it once, execute

programmable V100 GPUs on Summit,

slow wet-lab experiments.

it anywhere’ allows the same code to be

achieving similar speedups compared to

reused on different target devices.

the OpenCL version. In addition, with the

simulation, which searches for organic

However, few of the existing computational

help of Jubilee Development, OpenMP

docking tools were designed with high

AutoDock-GPU runs a hybrid search

was introduced to add another level of

performance computing in mind, so

combining a genetic algorithm with a

parallelization, namely the overlapping of

they need to be optimized in order to

local search phase. The implementation

input / output (I/O) to / from the GPU with

take full advantage of the computing

uses multi-level parallelization, which

the calculations, thus enabling efficient

resources available today. One example is

maps

virtual screening on truly large-scale runs.

AutoDock, an open-source software tool

methods, atomic forces) into suitable

for molecular docking simulation that was

OpenCL processing elements (kernels,

In

initially developed by Scripps Research

work-items). As a result, this OpenCL

docking simulations on the Summit

three decades ago. In recent years, an

version improved the processing speed by

supercomputer, the amazing speed-up

international effort has been made to

50 times compared to the original single-

witnessed by optimizing AutoDock for

docking

components

(search

addition

to

powering

molecular

GPUs has been put to use on computers all over the globe. This is thanks to recent efforts by Scripps Research and the World Community Grid (WCG) team, which have enabled the use of AutoDock-GPU in Leonardo Solis-Vasquez, Andreas F. Tillack, Diogo Santos-Martins, Andreas Koch and Stefano Forli 24 HiPEACINFO  64

the OpenPandemics COVID-19 project.

Image credit: WCG Open Pandemics

Peac performance


Peac performance The WCG initiative allows anyone with a computer and internet connection to donate their computer’s processing capabilities to contribute to solving largescale problems in health and sustainability. Participating

research

projects

agree

to openly share all the data generated with the community. To contribute, the volunteers download a secure software program

to

their

local

computers,

enabling them to perform AutoDock calculations in the background with their spare capacity. WCG collects the results from the volunteers, and combines them for forwarding to the Scripps Research

The SUMMIT supercomputer at Oak Ridge National Laboratory, one of the most powerful

team for analysis.

machines in the world © Oak Ridge National Laboratory, United States Department of Energy

For the OpenPandemics COVID-19 effort,

package of 10,000 ligand dockings, now

The AutoDock-GPU program is under active

volunteers can contribute both their

needing about two and a half GPU-days

support and development, and the code is

CPU and also their local GPU capacity,

as opposed to three CPU-years. Although

available under an open-source licence. In

if available. Due to the wide spectrum

individual high-performance GPUs can

addition to the GPU-based performance

of machines in WCG, which range from

complete these data packages an order

boost, AutoDock-GPU has been extended

Raspberry Pis to multicore servers, and

of magnitude faster (for example, an

with an improved search method that

GPUs ranging from integrated to high-end

NVIDIA Quadro RTX 6000 requires around

further increases the probability of finding

devices, typical calculation speedups of

eight hours), the above is an impressive

molecules

GPUs vs CPUs lie between 300-600 times.

speedup for a wildly heterogeneous mix of

Furthermore, ongoing porting efforts to a

with

antiviral

properties.

volunteer GPUs. Thanks to the hundreds

wider range of accelerator devices (such

The graph below shows the total speedup

of thousands of WCG volunteers, this

as vector-based computers and field-

histogram, with a real-world overall

grassroots approach is capable of providing

programmable gate arrays, or FPGAs)

GPU-CPU speedup of 454 times. This

computing power similar to that of leading

promise interesting insights from a high-

was achieved when processing an entire

large-scale HPC systems like Summit.

performance

computing

architecture

perspective. We invite researchers in the field to download, provide feedback, and even to contribute to the development of AutoDock-GPU.

FURTHER INFORMATION:

Code repository on GitHub: github.com/ccsb-scripps/AutoDock-GPU AutoDock-GPU’s main peer-reviewed article – please cite this when referring to the software: pubs.acs.org/doi/10.1021/acs.jctc.0c01006 Porting AutoDock-GPU to the Summit supercomputer: dl.acm.org/doi/ abs/10.1145/3388440.3412472 OpenPandemics - COVID-19 WCG effort: Speedup histogram of individual work units and overall package speedup (454x) of one package

bit.ly/WCG_COVID_GPU

(#30008, containing 10,000 ligand dockings), performed on the World Community Grid’s OpenPandemics COVID-19 project. The CPU runs were executed from 31 December 2020 to 7 January 2021, and the GPU runs in a beta test run on 31 March 2021 HiPEACINFO  64 25


SME snapshot

Even before the global COVID-19 pandemic, using advanced computing to accelerate drug discovery was a research topic of vital importance. Here, Chief Executive Ezequiel Mas de Molino introduces NOSTRUM BIODISCOVERY, a spin-off from the Institute for Research in Biomedicine (IRB) and Barcelona Supercomputing Center (BSC) that is making great strides in the pharmaceutical industry.

Fast-tracking pharmaceuticals with Nostrum Biodiscovery • Enzyme engineering: Torre Quevedo grant (no. PTQ2020011290) awarded by the Spanish Ministry of Science to Joan Coines (PhD) to design PluriZymes (enzymes with more than COMPANY: NOSTRUM BIODISCOVERY MAIN BUSINESS: drug discovery, enzyme engineering LOCATION: Barcelona WEBSITE:

nostrumbiodiscovery.com

Founded in 2015 by professors Victor Guallar (BSC) and Modesto Orozco (IRB), NOSTRUM BIODISCOVERY (NBD) is a spin-off that, despite its relative youth, has already made a major impact on the European biotech scene. The company specializes in early drug discovery / design and enzyme engineering, applying its expertise to help customers

one active site) over the next three years. • Nucleic acids: Torre Quevedo grant (no. PTQ2020-011291) awarded to Javier Iglesias (PhD) to develop and test a new technology targeting RNA, a growing trend in the pharmaceutical sector for addressing unmet medical needs. • Software platform: Industrial PhD (grant no. DIN2020011231) awarded by the Spanish Ministry of Science to NBD information technology (IT) manager Anna Gruszka to further develop the NBD software platform. This platform integrates the various technologies developed by NBD and has generated considerable interest among the drug discovery teams of our pharmaceutical clients.

accelerate and improve the efficiency of their pipelines. It has an excellent track record of securing collaborations with

Another recent triumph for the company was the signing of

pharmaceutical companies – including AstraZeneca and Almirall

a strategic alliance with Evoenzyme to create BioDiscovery

– as well as research institutes and some of the largest consumer

Enzymes (BDE). This partnership will combine the academic and

goods companies in the world.

industrial expertise of the two spin-offs in the field of enzyme evolution. By harnessing the knowledge-generation power of

Despite the challenges associated with the coronavirus pandemic,

computational and wet labs with proprietary technologies, BDE

the last 12 months have been particularly exciting for the

will enhance the properties of enzymes of industrial interest.

company: not only did it manage to break even, but the team was expanded significantly and the company moved to a new

At a time when, perhaps more than ever, biotech is in

office located within Barcelona Science Park.

serious demand, we are confident that NBD will contribute to biotechnology breakthroughs that will advance future

In June 2021, NOSTRUM BIODISCOVERY was awarded three

pharmaceutical developments, as well as providing greener

independent fellowships for team members in the areas of

strategies for industrial production thanks to the efficiency of its

enzyme engineering, nucleic acids and the Nostrum Biodiscovery

technologies.

software platform, as follows:

26 HiPEACINFO  64


SME snapshot With high-performance computing (HPC) providing the tools that are enabling scientific discoveries, companies like CVIS International, which provides a range of HPC services, are a huge help to scientists who just want to concentrate on their day job. Account Manager Gabriël Henriette gives us an introduction to CVIS.

Delivering the compute behind scientific breakthroughs technology partner providing expertise and services for scientific research and development (R+D). COMPANY: CVIS International

Founded in 2006, CVIS International is an established HPC

MAIN BUSINESS: high-performance computing (HPC) services

partner in R+D, providing HPC architecture design, hardware

LOCATION: Boxtel, Netherlands

delivery, implementation services, managed services and support

cvis.nl

for new and existing HPC clusters – including onsite HPC

WEBSITE:

CONTACT: Gabriël Henriette, Account Manager

clusters, cloud-based clusters and hybrid forms. With a strong network of technology partners including Microsoft and Bright Computing, CVIS has successfully implemented dozens of HPC clusters, many of which it still manages today. A recent example is the implementation of a large HPC production cluster for a university medical centre in the Netherlands conducting

Gabriël Henriette, Tiemo Kersten and Radjesh Angoelal, CVIS International

genetic research. Upon successful completion of the project, CVIS provided management services, and continues to manage the

What secrets does our human genome hold? Would it be possible

cluster to this day. The company is also a reliable strategic partner,

to create the perfect medicine or treatment and tailor it with

helping chart a roadmap for the future direction of the cluster.

our unique individual DNA for optimal effects? Human curiosity has often led to intriguing and complex questions like the ones

At CVIS, our main aim is to help R+D organizations shorten

above, many of which remain unanswered. Today’s state-of-the-

their ‘time to insight’, so they can accelerate scientific discoveries.

art methods to find answers to these questions include the use of

CVIS acts as a single point of contact for R+D organizations in

complex simulations and calculations. For example, a simulation

terms of the management of their HPC resources. This gives R+D

to predict the fold of a protein, or of the trajectory of fluids in the

professionals more time to focus on their core business and become

human body to determine where a particular medicine will end

more adept at using their HPC cluster, as opposed to spending time

up, or an analysis of the output of raw genetic sequencing data.

performing tasks that are not directly related to their profession,

Such simulations require an immense amount of compute power,

such as keeping the HPC infrastructure operational.

executed on a cluster computer. CVIS also helps researchers who are developing the technology for For scientists studying these fundamental research problems,

HPC. For example, new ideas for computer architecture, compilation

it is quite challenging to remain at the top of their scientific

and optimization need to be tested in a real world scenario. CVIS

field while at the same time staying up to date with the latest

assists these researchers by building HPC environments to their

developments and best practice in the field of HPC – a profession

desired test specification, so these researchers can validate their

in its own right. This is where CVIS International comes in, a

new concepts. What HPC challenge do you have for us? HiPEACINFO  64 27


In this issue we find out how SMARTLAGOON is developing a digital twin to enable real-time healthchecks for the Mar Menor, learn how SMART4ALL is powering new healthcare and energy technologies in Southeastern Europe, and get up to speed on HealthyCloud’s plans for a pan-European health cloud.

Innovation Europe REAL-TIME HEALTHCHECK FOR COASTAL LAGOONS: SMARTLAGOON In HiPEACinfo 58, we learned about

Ramsar Site – that is, on the List of Wetlands of International

technology

water-

Importance. We selected the Mar Menor as a testbed for the

based disasters developed by HiPEAC

SMARTLAGOON approach as it suffers from intensified pressures

Technology Transfer Award winner

– agriculture, tourism, fishing, urbanization, eutrophication

José M. Cecilia (Universitat Politècnica

(nutrient pollution), contamination, flash flooding, climate

de

partners.

change – that have led to environmental deterioration. Through

Building on that work, José’s team

negative feedback, these, in turn, have an impact on some of

solutions

València-UPV)

for

and

is embarking on a new project, SMARTLAGOON.

the socioeconomic uses.

Coordinated by the Universidad Católica San Antonio de Murcia

From a technological point of view, the Mar Menor poses

(UCAM) and funded by the European Union, SMARTLAGOON

challenges because of its large geographical spread (135km2),

will create a digital twin that will allow systemic understanding

which limits internet connectivity in many regions. This means

of coastal lagoons: uniquely vulnerable yet intensely rich

that new approaches for monitoring, modelling and forecasting

ecosystems. We caught up with José to find out more.

socio-environmental dynamics are needed, in order to restore, preserve and manage the lagoon in future.

Why did you choose to focus on coastal lagoons for this project? Coastal lagoons – shallow bodies of water separated from the

How will SMARTLAGOON help provide greater understanding

open sea by a barrier such as a reef – occupy 13% of the world’s

of coastal lagoons’ ecosystems?

coastlines, or around 5.3% of the coastlines in Europe. Among

SMARTLAGOON is developing a digital-twin strategy for

the most productive ecosystems on the planet, they offer rich

modelling and predicting socio-environmental processes across

opportunities for fishing and aquaculture, provide locations for

different temporal and spatial scales. This will allow citizens,

saltworks, and support a variety of recreational activities. In

researchers, stakeholders and policy makers to collect data in a

addition, they perform roles that are critical to the ecology of

more cost-effective way, and to create more precise models and

coastal areas, such as retaining and purifying pollutants.

predictions to support better decision making.

However, these natural systems are especially vulnerable to

This platform will be developed through co-design strategies

climatic and anthropogenic pressures. Moreover, they are

to raise public awareness of sustainability issues in a specific

classified as highly complex systems, thanks to their significant

environment (lagoons) through citizen science activities,

heterogeneity and the complexity of their hydrological,

including not only hands-on initiatives and remote crowd­

hydrodynamic, ecological and socioeconomic processes.

sourcing data-collection campaigns, but also dissemination events such as seminars, open days and science days, and open-

SMARTLAGOON focuses on the Mar Menor (Murcia, Spain), the

innovation initiatives. The outcome will be disseminated via

largest hypersaline coastal lagoon in Europe and a designated

data visualizations that allow interactive exploration.

28 HiPEACINFO  64


Innovation Europe

What are the main technologies to be used in SMARTLAGOON?

natural and social factors, and as a result of the involvement of

SMARTLAGOON will take an integrated approach that couples

relevant stakeholders, SMARTLAGOON will develop a forecast

novel intelligence-based sensing technologies as input data to

portal that will be able to simulate and predict short- and long-

an efficient internet of things (IoT) infrastructure, allowing

term policy scenarios.

innovative modelling of socio-environmental dynamics to forecast short- and long-term changes to the state of coastal

How will interested parties be involved in this project?

lagoons, and thus informing decisions to safeguard the

A key component in the management of bodies of water is that

ecosystem that coastal lagoons provide.

stakeholders and citizens have a sense of ownership of the problem and solutions. Stakeholders play a fundamental role

In terms of HiPEAC-related topics, SMARTLAGOON will combine

in SMARTLAGOON: they will be used to develop participatory

low-cost sensing technologies such as in-situ IoT infrastructure,

models and scenarios, data collection, and interpretation of

novel artificial intelligence (AI)-based sensing technologies

results.

through video monitoring, social sensing crowdsourced data, and satellite to measure socio-environmental variables on

From the outset of the project we have been raising awareness

a large scale. All of these will be developed with a reduced

among stakeholders and including them in project decision-

environmental footprint.

making and development through workshops and meetings. So far, these workshops have been held online due to COVID-19

As an example, the UPV is leading the task to design two

restrictions, but in-person events should be possible in the future.

efficient IoT infrastructures that will be deployed in the Mar Menor lagoon and the Segura watershed. These will be powered

We also plan to develop citizen-science initiatives in order to

by connectivity technologies and architectures such as LoRa or

open up knowledge to the general public. Of special interest

SigFox to provide the data in near real time with a low power

to us are schools: secondary-school students (in this case 14

budget. Edge / fog computing capabilities will also be enabled,

to 18-year-olds) are digital natives. Most have a mobile phone

with computational horsepower being introduced at the

and can download the DischargeAPP application, developed

edge thanks to graphics processing unit (GPU) virtualization

by SMARTLAGOON partner Photrack, to upload videos of

techniques such as rCUDA.

ephemeral streams so we can determine the input flow draining into the Mar Menor. They are also very active on social media,

SMARTLAGOON will also research, demonstrate and test a

giving us sufficient critical mass for our social sensing strategy.

novel multi-model approach using not only physical models

Not to mention, of course, the benefit to society of making new

but also machine-learning techniques, a combination which

generations aware of the serious environmental problems our

represents a great challenge for today’s scientific community.

society is suffering on the one hand, but the great solutions that

Based on these innovative modelling approaches, combining

technology can offer on the other.

The Mar Menor lagoon in Murcia has been chosen as the site for the project HiPEACINFO  64 29


Innovation Europe

When do you expect to have initial results and how will we

developments and join us in using scientific advances for the

be able to view them?

protection of our environment.

SMARTLAGOON is a very ambitious project that will deliver a number of science-to-technology breakthroughs, varying in

SMARTLAGOON has received funding from the European Union’s

maturity from technology readiness level (TRL) 3 to TRL 8.

Horizon2020 research and innovation programme under grant

The project started on 1 January 2021, and, like many of our

agreement no. 101017861.

colleagues, the consortium is working hard to meet the schedule set out in the grant agreement. The first technical deliverables will be available from June 2022, when we expect to have the monitoring system fully deployed in the Mar Menor area. We warmly invite the HiPEAC community to follow project

FURTHER INFORMATION:

smartlagoon.eu  @SMARTLAGOON

SMARTER ENERGY METERING AND SOCIAL DISTANCING PRECAUTIONS, THANKS TO SMART4ALL In HiPEACinfo 61, we learned about SMART4ALL, a Euro­

minimum distances between users using the sound propagation

pean-funded ‘marketplace-as-a-service’ Innovation Action

time, which is also robust against ambient noise thanks to

funding technology transfer experiments in South­ east

modern signal processing methods.’

Europe, as well as providing tools, services and platforms. A year and a half into the project, SMART4ALL Communi­

Warning signals – in the form of vibrations, bright LED flashes

cations Officer Katerina Labrakopoulou (Patras Science

and/or acoustic cues, are enabled by the bracelet’s technology

Park) presents some of the new technologies which have

when the user gets too close to others, a distance that can be

been enabled thanks to the project.

configured from one to three metres. ‘All data processing is

EmBRACE smart social distancing bracelets

done on the devices, and no data is logged or shared, leading to enhanced security. The devices are battery powered, offering activity times similar to those of a smart watch, and use a

Prior to COVID-19, the cruise industry brought in around

standard USB charging port,’ explains Johannes. ‘You can

€34 billion annually in Europe. However, few will forget reports

disable the devices by bringing them very close to each other,

of passengers becoming stranded as the pandemic took hold.

so they don’t need a switch. They can be configured through

Can innovative technology solutions help ensure passenger

an easy-to-use mobile app, for example for creating a group of

wellbeing and restore confidence in a post-COVID world?

several family members, disabling the warning signal among the members. They can also be connected to an infrastructure

Ioannis Kostopoulos of Greek systems integrator Optionsnet

such as a base station to enable emergency evacuation support.’

and Johannes Wendeberg of German indoor sensing company Telocate certainly think so. Their EmBRACE project, which received €80,000 in funding from SMART4ALL, sought to respond to the urgent demand for safety measures in cruise and passenger shipping. ‘Maximum density policies for shared spaces are one way of preventing infectious outbreaks, and monitoring technologies provide a way of ensuring compliance with these,’ explains Ioannis. ‘In this regard, positioning technologies provide a way to help manage future pandemics.’ EmBRACE has developed a social distancing support bracelet to support compliance with distancing regulations. ‘In the past, solutions of this kind relied on radio frequency signals,

The EmBRACE bracelet provides a reminder about social distancing.

which are often imprecise and may lead to user irritation,’

Ultrasound and Bluetooth ensure precise, reliable distance

says Johannes. ‘By complementing Bluetooth with ultrasound,

measurements, and violations of distancing rules are indicated by

EmBRACE provides a far more precise way of measuring

vibration and flashing

30 HiPEACINFO  64


Innovation Europe

The target market for the EmBRACE solution comprises over

The results showed that a simplified architecture would fit

1,300 large passenger ships or ferries in addition to thousands of

the smart meter, keeping memory requirements low and the

smaller vessels, which together transport more than 2.1 billion

runtime in the range of milliseconds. ‘As the aggregated load

passengers, 250 million vehicles and 32 million trailers every

of a building remains stable when no appliance enters or exits

year, not to mention over 400 cruise ships. However, EmBRACE

the electricity network, the estimation of energy consumption

could also be used in other settings, including companies or

is only useful when a significant change is detected in the active

schools.

power,’ adds Iosif. ‘Thus, the proposed scheme suggests that the integrated algorithm is capable of estimating the power of

‘Thanks to SMART4ALL, we’ve managed to combine Telocate’s

each appliance whenever an event occurs in the aggregated

sensing expertise with Optionsnet’s knowhow in systems

power. The last phase of the project will focus on optimizing the

integration,’ commented Ioannis. ‘This is helping us deliver

algorithm’s accuracy by minimizing the energy consumption

innovative services that will improve passenger and crew safety

estimation error, in order to deliver a sustainable solution that

by monitoring and analysing their behaviour through smart

meets the market’s needs.’

wearables.’

Energy-saving smart meters, thanks to disaggregation

As a stand-out example of technology transfer, EDIoT, while having a strong research aspect, will result in a real product that is expected to make a real impact in the field of energy

Smart meters have become increasingly prevalent and provide

monitoring, providing a powerful tool for non-intrusive

a key tool in the management of energy resources. ‘By solving a

appliance detection that in turn aims to enhance energy

problem known as energy disaggregation (see figure 1), energy

efficiency. ‘By bringing together a lightweight, robust algorithm

meters allow users to see an estimate of how much energy is

in a low-cost yet powerful smart meter, Meazon is pioneering

consumed by each appliance,’ explains Aimilia Papagiannaki

machine-learning techniques for energy monitoring solutions,’

at Meazon, a Greek energy management solutions company.

says Aimilia. ‘We look forward to building on this work for

‘However, until now solutions for energy disaggregation have

future product roll-outs.’

been executed in the cloud, meaning that high-frequency data has to be transmitted from the meter to the cloud, which is costly in terms of resources.’ As part of the ‘Energy Disaggregation on IoT Smart Meters’ (EDIoT) project funded by SMART4ALL, Meazon teamed up with the University of Hertfordshire in the UK to integrate an energy disaggregation algorithm within a state-of-the-art smart meter. ‘We developed a submeter able to sample every 10msec,

Energy disaggregation concept

providing high-frequency measurements which in turn help the development of more accurate algorithms, as shown in figure 2,’ says Aimilia. ‘Implementing this technology within the smart meter means that the heavy processing is performed at the hardware level and thus minimizes the resources needed to transmit data to a remote server.’ To achieve this, the partners used machine learning techniques. ‘We developed a light version of a high accuracy neural network architecture that was able to meet the constraints of the hardware,’ notes Iosif Mporas of the University of Hertfordshire. ‘A thorough examination of the parameters was undertaken to define the system’s limitations and determine the most

Meazon’s high-frequency smart meter

appropriate architecture. Different schemes were deployed to establish the network size, using features from both the time

SMART4ALL has received funding from the European Union's Horizon

and frequency domains.’

2020 research and innovation programme under grant agreement no. 872614

HiPEACINFO  64 31


Innovation Europe

HEALTHYCLOUD: DEFINING THE STRATEGIC AGENDA FOR THE EUROPEAN HEALTH RESEARCH AND INNOVATION CLOUD Adopting best practice in health data management is essential

The project is driven by two real-world use cases on cancer

in order to make significant advances in health and biomedical

and atrial fibrillation, which will ensure that propositions by

sciences. Accordingly, one of the priorities of the European

domain-specific and technological experts are technically and

Commission 2019-2025 is the creation of a European Health

ethically sound and legally compliant. The ultimate goal is to

Data Space (EHDS), which will be a means to improve health

draft the blueprint of an ecosystem that builds and reinforces

research and its translation to healthcare at all levels: from

the trust of patients and citizens in the use of their health data

public health to personalized medicine, through secondary use

for research, actionable through a portal that serves as an

of health data.

interface with the cloud services.

HealthyCloud will deliver a strategic agenda for the implementation of the European Health Research and Innovation Cloud (HRIC), which will be one of the cornerstones of the EHDS. The agenda will incorporate consolidated feedback from a broad range of stakeholders, including the European Commission, European Union (EU) member states and regional, national, European and other relevant international initiatives. Towards the European Health Research and Innovation Cloud (HRIC) ‘In the HealthyCloud project, we have joined forces from multiple health and health-related domains to shape the future Health Research and Innovation Cloud,’ said Juan GonzálezGarcía, Biocomputing Unit manager at the Institute for Health Sciences in Aragon (IACS) and HealthyCloud co-coordinator. ‘HealthyCloud will pave the way towards effective healthrelated data sharing across Europe, contributing to the realization of the European Health Data Space. It is important to understand existing barriers that limit the effective exchange of health-related data. This effort will allow us to propose ethically sound, legally compliant and technically feasible open standards and protocols to overcome them,’ said Salvador Capella, Spanish National Bioinformatics Institute (INB) coordination node leader at Barcelona Supercomputing Center (BSC) and HealthyCloud co-coordinator. HealthyCloud is organised around four fundamental objectives that cover: • interactions with stakeholders to ensure their voices are included as part of the strategic agenda • inclusion of ethical, legal and societal aspects in the design of the future HRIC ecosystem • sustainable access, use and re-use of health-related data,

PROJECT NAME: HealthyCloud: Health Research & Innovation Cloud START/END DATE: 01/03/2021 – 31/08/2023 KEY THEMES: health data, health research and Innovation cloud,

European Health Data Space, big data, cloud computing PARTNERS: Spain: Instituto Aragonés De Ciencias De La Salud

(IACS) (co-coordinator), Barcelona Supercomputing Center (BSC) (co-coordinator), Universitat De Barcelona (UB), Servicio Andaluz De Salud, European Genome-Phenome Archive (EGA); Austria: Biobanks And Biomolecular Resources Research Infrastructure Consortium (BBMRI-ERIC), Austrian National Public Health Institute (GÖG); Belgium: Sciensano; Estonia: University of Tartu; Finland: Euro-BioImaging, Finnish Institute for Health and Welfare, CSC - IT Center for Science; France: European Clinical Research Infrastructure Network (ECRIN); Germany: Technologie Und Methodenplattform Fur Die Vernetzte Medizinische Forschung Ev (TMF), German Network for Bioinformatics Infrastructure (de.NBI); Luxembourg: University of Luxembourg; The Netherlands: EGI Foundation, European Infrastructure for Translational Medicine (EATRIS-ERIC), Leiden University Medical Center; Portugal: Serviços Partilhados Do Ministério Da Saúde Epe (SPMS); United Kingdom: ELIXIR Europe BUDGET: €2,999,975

available in several data collections spread in data hubs, considering a progressive adoption of the FAIR – findable, accessible, interoperable and reusable – principles • technological solutions in terms of computational facilities and mechanisms to enable distributed health data analysis across Europe 32 HiPEACINFO  64

healthycloud.eu  @HealthyCloudEU  HealthyCloudEU


Technology opinion With quantum computers still a way off, quantum may not be a priority for many academic or industrial researchers. This is a mistake, argue Koen Bertels, Aritra Sarkar and Imran Ashraf (QBeeX); given the likely enormous impact of quantum on a huge range of areas, they urge universities and companies to start incorporating quantum into their research now.

Don’t wait for quantum hardware to mature

Start researching quantum logic and algorithms now, based on perfect qubits them to make certain kind of computations. Thanks to these properties, quantum computing represents a game changer, with substantial speed-ups over today’s machines foreseen. However, there have been a number of inflated claims and results that are very difficult to prove or even replicate. In fact, realistically, we are still at least 10 years away from a functional quantum Left to right: Koen Bertels, Aritra Sarkar, Imran Ashraf

computer.

The state of quantum: First, the physics

Recently, we published a paper in IEEE Micro (see ´Further

It was a talk by American physicist Richard Feynman in 1981

reading’, overleaf) to clarify the status of quantum computing

that first prompted mass-scale adoption of quantum computing.

research. The long-term goal is to create a quantum chip

The talk introduced mysterious quantum phenomena such as

incorporating a high number of good-quality qubits, which can

superposition, which, along with other phenomena, we still do

be implemented using quantum error correction; this is known as

not fully understand, although we have found ways to induce

‘fault-tolerant quantum computing’ (FTQC). However, there are a

Full-stack quantum computing architecture and qubit technologies HiPEACINFO  64 33


Technology opinion number of fundamental physics-based challenges to be resolved

This PISQ approach would allow universities but also industrial

before this can be achieved. For example, there is still no clear

companies to start bringing quantum research into any scientific

frontrunner in the technologies that could be used to make the

field. As an example, look at genomics, where we have defined

best qubit, although the Majorana approach seems to be falling

the quantum algorithms necessarily to make functionally similar

off and we would expect other technologies to converge to a

genomics sequencing and analysis computations on a quantum

dominant approach over the next few years. [See HiPEACinfo 56

device. Quantum genetics research has achieved many results, of

pp.24-25 for a summary of qubit technology candidates].

which the new quantum gates are perhaps the most important as they offer new ways to implement the genome sequencing or

Our paper outlines a list of open problems relating to physical

genome analysis algorithms.

qubits that can be solved by researchers with a quantum physics background. A promising direction in this area is ‘noisy

Quantum chemistry is another major area globally – with

intermediate-scale quantum computing’, or NISQ, where ‘noisy’

applications within the automotive sector, construction and

refers to gate errors and coherence observed in the physical

pharmaceuticals, to name a few – and at QBeeX we have started

qubits, and ‘intermediate-scale’ refers to the number of physical

a new line of research in this area with the University of Leuven,

qubits.

which includes working with a large chemical company in the

Advancing quantum research using the full stack

Port of Antwerp.

Beyond this, however, there are many other aspects that need

Leading new quantum directions at QBeeX and Leuven

intensive, long-term research activities. Quantum computing

QBeeX, the company we have recently launched, is closely

represents a revolution in the way computations are made;

connected to the University of Leuven, allowing us to offer PhD

consequently, concepts, theories and applications – in fact,

opportunities to our employees. Its main goal is to develop a

anything in science – will have to be redeveloped from scratch.

quantum accelerator for large industrial companies in Europe.

This means that every university will need to initiate quantum

We strongly believe that universities across the world need

research in all the scientific fields they cover, while companies

to allow young academic researchers to develop a profound

should get quantum researchers working on current problems in

understanding and a good understanding in order to perform

their sector.

long-term quantum research for scientific fields relevant to the university. Training people to advance investigations into the

For us, it is evident that the first use of any quantum algorithm

influence of quantum on their respective fields will allow us

will be on a quantum accelerator. In our research collaboration

to deliver the maximum benefit from quantum computing as a

between TU Delft and Intel, we developed a full-stack quantum

scientific tool.

computing architecture for such an accelerator, from the application to the microarchitecture and a quantum simulator.

FURTHER READING:

These tools are freely available in the public domain. Once a physical quantum chip becomes available, we have the experience

Feynman, R. ‘Simulating Physics with Computers’

to move the simulator to a physical implementation.

International Journal of Theoretical Physics volume 21, pages 467-488 (1982)

In the meantime, we propose a new approach which would allow

bit.ly/IJTS_Feynman_1981

any scientific field to incorporate quantum into its research. This is ‘perfect intermediate-scale quantum computing’, or PISQ.

Bertels, K., Sarkar, A., and Ashraf, I. ‘Quantum Computing - from NISQ

‘Perfect qubits’ are a theoretical concept, where the qubits do

to PISQ’

not decohere to a ground state in am extremely short time, and

IEEE Micro, September/October 2021

where correct quantum gates are performed on those qubits. Even

bit.ly/PISQ_IEEE_Micro_Sept_Oct_2021

companies like IBM now compile also for perfect qubits to develop and test quantum algorithms expressed in perfect qubits. Our proposal is that application and algorithm developers working

Interview with Koen Bertels in HiPEACinfo 56, pp.24-25 bit.ly/HiPEACinfo56_Koen_Bertels

on the highest layer of the full stack express their concepts in terms of perfect qubits, which could be simulated on classical

Gamrat, C. ‘Towards operational quantum computing? Or, thinking

supercomputers. The main cost is the amount of memory needed

beyond qubits’

to represent the data of superposed or fully entangled qubits,

HiPEAC Vision 2021

currently limited to around 50-60 perfect qubits. 34 HiPEACINFO  64

bit.ly/HiPEACVision2021_quantum


Technology opinion

How will computer architectures respond to the ever-increasing demands of artificial intelligence (AI) workloads? In this article, Kemal A. Delic, visiting fellow at the Open University (UK) and independent technology specialist Martin Antony Walker set out their vision of a computer architecture capable of powering the AI-enabled scientific advances of the future.

Architecture of large-scale AI systems In brief, a division of AI architecture

In an analogy with the functioning of

into computation-heavy and cognition-

the human brain, we outline a hybrid

intensive parts will occur, orchestrated

architecture divided into computation-

into a unified whole.

heavy

cognition-intensive

parts

(Figure 1). While the first is already present

Hybrid AI architecture

in large data centres, we believe that the

Major advances today are made by

Kemal A. Delic, Martin Antony Walker

and

second will appear in the next few decades.

institutions deploying very large compu­ Science is an endless enquiry into the

ta­ tional

vast

While GPUs were initially created for

mysteries of nature, of matter and of life.

data collections to enable insights into

the parallel processing of images, they

Advances are made through scientific

profound

Drug

also prove useful for heavy numerical

discoveries in which scientific instruments

discovery, gravitational-wave astronomy,

calculations, since matrix manipulations

play an important role. We are witnessing

the structure of biological molecules,

underlie many AI tasks and workflows.

an explosion of interest in AI-inspired

and investigations of subatomic particles

Thus we denote the CPU-GPU part of

methods to enable discoveries. We believe

are just a few examples of such uses.

the hybrid architecture of Figure 1 as

the architecture of large-scale systems will

The deployment of graphics processing

‘computation-heavy’.

be very different in the next few decades

units (GPUs) in large data centres is the

and will certainly reflect the necessity

current dominant trend of accelerated

The current development of a number

of addressing hugely complex problems

computation.

of neuromorphic chips points towards a

infrastructure mysteries

in

accessing nature.

in the life and physical sciences. We sketch here the architecture of a hybrid, large-scale system, and illustrate the required flow of problem decomposition, scheduling and execution. Recent major advances in AI have been enabled power

by driven

increased by

compu­ tational

increased

central

processing unit (CPU) clock frequency, the introduction of multicore chips, and increased exploitation of parallelism. In this short article we claim that for the

© Kemal A. Delic - Martin A. Walker - 31 July 2021

next stage of improvements a different hybrid AI architecture will be necessary.

Figure 1: Architecture of a large-scale AI system HiPEACINFO  64 35


Technology opinion future architecture capable of addressing

of the problem into fragments fitting into

address previously intractable problems

cognition-intensive problems employing

each architecture segment. It will also

and resolve some great challenges. But

neuromorphic computing units (NCUs).

require some very inventive steps, and will

this will take time, it will need passion and

These

be essential for successful deployment.

commitment, and it will require resources.

will

work

in

harmony

with

quantum computing units (QCUs), which will be used for specialized simulations

Problem-workflow mapping Outlook

and enhanced security. Development of

Thinking about these large facilities, able to

Large-scale infrastructure of the kind

this cognition-intensive part will take a

deal with singular, challenging problems,

described in this article will constitute

longer time, and will require significant

will always require transformation of the

scientific

instruments

investment and prolonged research. An

problem at hand into a model suitable

important

advances

interesting

fundamental

for execution, the model comprising

scientific breakthroughs. Such instruments

research in quantum biophysics has been

workflow and workloads which will be

represent a basis for doing ‘big science’.

suggested by a recent physics Nobel prize

scheduled for execution and monitoring

The focus will certainly be on scale,

winner, who proposed a ‘quantum mind’

(figure 2).

speed, and energy consumption, and the

direction

of

as a potential scientific breakthrough.

that and

enable potential

ultimate result will be that we know more, As important parts of the architecture

learn faster, and understand better. This

The key to a functioning hybrid archi­

described here are still in active develop­

has always been the result of the use of

tecture is represented in the middle part

ment, we believe that experimentation

advanced instruments in the past – as

of figure 1, providing brokering, data

and the art of problem decomposition

the history of microscopes and telescopes

flows and control execution between and

and execution require further exploration

convincingly demonstrates.

among principal functional components.

which may indicate a path to big advances.

This element will depend very much on

The winners will learn how to deploy this

the problem domain, and a decomposition

advanced scientific instrumentation to

© Kemal A. Delic - Martin A. Walker - 31 July 2021

Figure 2: Flow of problem decomposition, scheduling and execution

36 HiPEACINFO  64


HiPEAC futures At HiPEAC, we know that students setting out on their research careers are the future of computing systems. That’s also the view of the DATE conference, whose Young People Program is specifically designed to attract early career researchers. Anton Klotz (Cadence) tells us more.

Electrifying futures: The DATE Young People Program The DATE conference is the largest electronic design and automation (EDA) conference in Europe. DATE stands for Design, Automation and Test in Europe, although the ‘E’ is interpreted as ‘Embedded’ when it comes to the topic areas for submission. One focus of the conference is to integrate the new generation

“Finally, DATE got younger! By providing PhD students a new form of discussion, an interactive exchange of ideas was possible even in an online format” Georg Gläser; Mixed-Signal Engineer, Institut für Mikroelektronik- und Mechatronik-Systeme

of EDA researchers into the community by organizing engaging events. Last year, these activities were organized under the

experts who have taken different paths. This will give attendees

umbrella of the Young People Program. This was so successful

the opportunity to find out the benefits of different options and

that in 2022 the Young People Program will once again take

get an idea of which would be the best fit for them.

place in a virtual format, allowing young talent from all over the globe to take part. HiPEAC is working with the Young People

The Young People Program at DATE also hosts a PhD forum,

Program at DATE to help organize events as part of the overall

where PhD students present their research results in a poster

conference, which takes place on 14-23 March 2022.

session in front of an audience, followed by a discussion – which might even turn into an interview with a recruiter. Submit your

“The Young People Program of DATE is an excellent add-on to

proposal via the link below.

the scientific conference, as it allows the next generation of talent to learn more about the recruitment process and to get

Tutorials at DATE are a way for young talent to listen to experts

connected to industry representatives. We are looking forward

on subjects that might be outside of their current research

meeting students and graduates at DATE 2022”

scope, and will broaden your horizon with new perspectives.

Patrick R. Haspel, Global Program Director, Synopsys

Some tutorials include hands-on exercises, thanks to the cloud technology providing access to the tools.

The programme for 2022 will include a careers fair, providing a matchmaking opportunity for young talent with EDA and microelectronic design companies, so they can ask recruiters about career paths in industry. In parallel, at the university booth, research institutes and universities will display open positions for those who decide to work on new research challenges by continuing their academic career. Meanwhile, if you haven’t yet decided what your next career move should be, HiPEAC is organizing a careers forum with “The Young People Program at the DATE conference is a great opportunity for industry representatives to meet young researchers and offers PhD students a great overview of their career paths” Tal Zigman, Head of Global University Program, Cadence Design Systems

And if you thought that virtual conferences were unidirectional, with little audience feedback, you haven’t participated in a BarCamp event. BarCamp introduced the concept of a ‘particibutor’: a participant who has to contribute. Paticibutors pitch discussion topics and vote on topics to decide the agenda. Only one slide is allowed for each topic, so the rest is discussion, brain­ storming, blogging, etc. After its success in 2021, BarCamp@DATE will be returning in 2022.  date-conference.com/young-people-program Got a question about the Young People Program? Contact us:  young-people-program@date-conference.com Register for DATE via the conference website: Submit your proposal for the PhD forum:

date-conference.com

bit.ly/DATE_PhD_Forum

HiPEACINFO  64 37


HiPEAC futures Thinking about applying for an internship? You might think that going for a major international company is the best option – after all, a big name would look good on your CV. However, an internship at a smaller company can give you ownership of a project and be immensely satisfying, as Erhan Baturay Onural tells us.

HiPEAC internships: your career starts here Optimizing vision-based embedded devices at Sundance

Minimizing power usage in computer vision edge devices

NAME: Erhan Baturay Onural

With more and more edge devices employing computer vision,

RESEARCH CENTRE: TOBB University of

it is increasingly important to minimize power usage while

Economics and Technology

maintaining the required throughput. During my internship,

HOST COMPANY: Sundance Multiprocessor

I monitored power dissipation on a running application using

Technology

Lynsyn.

DATE OF INTERNSHIP:

To demonstrate this, I accelerated a simple matrix multiplication

01.03.2021 – 31.05.2021

– a function on which vision-based applications rely heavily becoming

– using Xilinx SDx. The application, which runs both on the

increasingly popular for connected embedded devices, currently

programmable logic and processor system, was executed on

numbering 20 billion and counting. FPGAs offer hardware-time

the VCS-1 with the Lynsyn application running in parallel

speed and reliability, crucial factors for the correct functioning of

on the Lysnsyn Lite board. As shown in the power samples,

real-time operating systems.

the application uses around 2.5W for both the programmable

Field-programmable

gate

arrays

(FPGAs)

are

logic and the processor system, a figure which is acceptable in In contrast to standard microcontrollers, they offer both

comparison with other edge devices.

hardware and firmware reprogrammability, parallel processing, lower power use and higher speed, as the component complexity

In terms of power-per-watt, FPGAs far outperform their graphics

is much lower. However, even this lower power use is still

processing unit (GPU) counterparts, making them more suitable

problematic: besides being a major consideration for battery-

for edge applications, where power saving is a crucial deployment

powered systems, it also matters for heat dissipation inside the

factor. Power tracking will be a key tool to enable more power-

closed environment in which embedded devices tend to operate.

efficient designs in the future.

VCS and Lynsyn Created by Sundance Multiprocessor Technology, the VCS-1 is a PC/104 Linux stack comprising an EMC2 board – a carrier for a system-on-chip (SoC) module – and an FM191 expansion card which fans out the inputs / outputs (I/Os) from the SoC to the outside world. The SoC – which can either be a Xilinx Zynq 7 Series or a Xilinx MPSoC Zynq Ultrascale+ – uses ARMbased processing to run the Linux operating system and memory interfaces, and programmable logic for hardware acceleration and general-purpose I/O. Sundance has also created the Lynsyn power utility board to measure the power usage of a system and correlate power values with the source code of the program running. There is also a ‘Lynsyn Lite’ version, which offers the same capabilities but with fewer measuring channels. 38 HiPEACINFO  64

The Sundance Lynsyn Lite board being used to measure power usage


HiPEAC futures

LynSyn output (power)

Flemming Christensen, Managing Director of Sundance Multiprocessor Technology, said: ‘HiPEAC’s internship has been very beneficial for Sundance to test new products or ideas on a “fresh” student. The benefit for the student is getting a feeling for working in the industry and learning about our edge artificial intelligence (AI) technology that is a direct result of the H2020 TULIPP project. We hope the programme will continue for many years.’

The ‘Power Profiling custom application on VCS-1’ project by Baturay and Ivica Matić , an embedded software engineer at Sundance, made it to the final of the 2021 Xilinx Open Hardware design competition. Congratulations on behalf of HiPEAC! TULIPP project website:

tulipp.eu

Towards Ubiquitous Low-power Image Processing Platforms:

springer.com/gp/

book/9783030535315 Finalists in the Xilinx Open Hardware design

Interested on high-impact vision applications on a low power

competition

openh w.eu/2021

budget? The Towards Ubiquitous Low-power Image Processing Platforms (eds. Magnus Jahre, Diana Göhringer, Philippe Millet) book is now available from Springer, synthesizing the knowledge

Video of the ‘Power Profiling custom application on VCS-1’ project  youtu.be/TrdNMubUr0s

gained by the consortium of the European project TULIPP.

The TULIPP project team at HiPEAC19 HiPEACINFO  64 39



Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.