2 minute read

SPINNAKER2

THE NATURALLY INSPIRED ARTIFICIAL INTELLIGENCE OF THE SPINNAKER2 SUPERCOMPUTER

A SUPERCOMPUTER THAT SIMULATES THE WAY THE HUMAN BRAIN WORKS

THE INNOVATOR: CHRISTIAN MAYR HALF­SCIENTIST, HALF­ENGINEER

Prof. Christian Mayr is an ordinary professor of highly concurrent VLSI systems and neuromicroelectronics at the Technical University of Dresden. He holds ordinary lectures, reads papers and organizes all things academic. There is an extraordinary side to Mayr, too, though. He sees himself as an intermediary between the higher education ivory tower and cool-running industry. His ideal: transferring scientifically excellent ideas to excellent engineering work. Sometimes, these kinds of comprehensive ideas also come to him during his limited free time, when he is tinkering around with circuits, microcontrollers and other controllers for the house (preferably together with his young son), which he built in the Dresden suburb of Radebeul.

A Supercomputer That Simulates The Way The Human Brain Works

One of these ideas is SpiNNaker2, part of the Human Brain Project that represents a simulation model of the human brain. This means that, for starters, it is actually a chip containing accelerators that can simulate neurons and synapses. You could also say the chip is a naturally inspired artificial intelligence network. From this single chip, Mayr and his team are building a 70,000-chip machine called SpiNN-cloud, which is distributed across 16 server racks. It is quite a sizable thing, both in thought and in space.

Double The Innovative Disruption

In June of 2021, the initial prototypes of the SpiNNaker2 chip were produced and put through their paces, both electrically and functionally. Based on the insight gained, the prototype version of the SpiNNaker2 chip was optimized and ultimately produced in great quantities on large wafers 300 millimeters in diameter by Globalfoundries in Dresden, Germany, for SpiNN-cloud.

What is especially fascinating about SpiNNaker2 is that it causes innovative disruptions in two areas at once: energy efficiency and the speed of information processing. According to Mayr, whether you are considering current versions of autonomous driving or robotics applications, for example, both still require far too much energy. In fact, about 100 times as much as it has to be and is allowed to be. This is because there is far too much calculation of unnecessary things—so-called redundancies. That is why it makes sense to use the human brain as a model, as it works according to the principle of information processing steps always preceding “communication”.

The information relevant for a problem that is actually transferred—and which consumes energy accordingly—is decided upon in these steps. What decides on energy consumption is therefore the process that takes place prior to “communication”. It is precisely this principle that Christian Mayr, who describes himself as being “heavily biology-inspired,” applied to the entire SpiNNaker2 system. That is why this supercomputer, which brings together artificial intelligence and brain biology, is far superior to ‘classical AI.’

“In the first step, we want to put a new generation of chips on the market that directly enable language processing, including full language analysis. In this way—that is, without a detour through the cloud—SpiNNaker2 not only saves time, but also enables better data protection in the case of ‘always-on’ applications,” said Christian Mayr. SpiNNaker2 is much faster at information processing, consumes considerably less energy and will ensure that people, machines and systems can interact much more efficiently in the future— including in medical technology, autonomous driving and robotics, for example. This is why Christian Mayr’s team is already working on the next SpiNNaker generation.

“In the future, one of our chips could be built into every robot joint, making a detour through a central computer obsolete in many cases. This will considerably improve training of the robot’s own kinematics.”

Development of the SpiNNaker2 project has continued since the summer of 2021 and completion of SPRIND-commissioned validation by SpiNNcloud Systems GmbH, which was co-founded by Prof. Mayr. SPRIND still maintains contact with the company on a regular basis.

This article is from: