-
RESEARCH
Pioneers of a new computing era
RESEARCH
Pioneers of a new computing era
The brain as a model: neuromorphic computers rely on an architecture that works in a similar way to the nerve cell network in our organ of thinking. John Paul Strachan and Emre Neftci have come to Jülich to make such computers ready for practical use.
Our brain is a masterpiece of evolution. In a gigantic network, 86 billion nerve cells are connected to each other via synapses and perform astonishing mental activities. John Paul Strachan and Emre Neftci both get excited when talking about the human brain and its incredible complexity, extraordinary capabilities and unprecedented efficiency. The two physicists aim to build computers that perform calculations based on the model of the brain, so-called neuromorphic computers.
Transferring the capabilities of the human brain to computers sounds tempting. “The neural networks in our brain process and store huge amounts of information. In contrast to computers, they spend very little energy doing this. They also keep flexibly adapting to new learning processes and experiences,” Neftci describes. Technical systems with such capabilities would be ideal, for example to navigate cars autonomously through traffic or for computers to learn by themselves.
Swiss-American Prof. Emre Neftci focuses on the software needed for neuromorphic chips. Most recently, he worked as an assistant professor at the University of California at Irvine. Before that, Neftci had completed his doctoral degree and did research in neuroinformatics at ETH Zurich. Although the first neuromorphic model circuits already exist, they have not been widely used in real-world solutions. Strachan and Neftci intend to change that. To this end, the two came to Jülich from the US high-tech region of California in 2021. “Jülich has experts from different fields who we need for this kind of interdisciplinary research,” says Neftci. He and Strachan complement the Jülich team of hardware and software engineers, semiconductor experts, basic scientists, theorists and neuroscientists.
However, building a computer that works in a similar way to our brain requires special components, a dedicated architecture and new algorithms. Unlike our brain, computers process and store data in different places. It takes time and consumes energy to send it back and forth between processor and memory. John Paul Strachan and his team are working on chips that process and store data simultaneously in one place. These so-called neuromorphic chips are modelled on the structure of neural networks in the brain and consist of highly interconnected artificial neurons and synapses.
Born in Costa Rica, US-American Prof. John Paul Strachan worked in Silicon Valley at Hewlett Packard Laboratories, one of the pioneers in the field of neuromorphic computing. Strachan led a team there researching neuromorphic hardware. He holds more than 50 patents and completed his degree at two of the elite American universities, MIT and Stanford. At the same time, the researchers are developing new concepts for the architecture of neuromorphic computers, for which they are designing additional devices and suitable circuits. Computers of this kind are to use stored data to further develop their own capabilities. Like our brain, they build and optimize internal information highways. Simply put: they learn. In this way, computers are to solve tasks for which they were not initially programmed.
New AI concepts
“Neuromorphic technologies could also fundamentally change artificial intelligence, or AI for short,” says Emre Neftci. Previous AI concepts have used mathematical algorithms without regards for an efficient physical realization. This could change with neuromorphic computers, as these are designed similarly to their natural – very efficient – counterparts.
However, new hardware and new AI concepts also require suitable software and algorithms. This is where software expert Neftci and his team come in. “The algorithms used for conventional computing cannot simply be transferred to neuromorphic computing. Different principles and restrictions apply there,” he describes the challenge. In order to realize suitable programmes, he and his team work closely with other Jülich researchers, such as those working in brain research. “Their findings can help us to further develop software for machine learning,” says Neftci.
0million euros
is being invested by the Federal Ministry of Education and Research in the development of hardware and algorithms for computers modelled on the brain. The ministry is funding the second phase of the interdisciplinary NEUROTEC project, in which John Paul Strachan and Emre Neftci are also involved. Coordinated by Forschungszentrum Jülich, NEUROTEC combines the expertise of Jülich, RWTH Aachen University and numerous regional companies. The partners want to help ensure that future computers for artificial intelligence will increasingly come from Germany and the EU.
Strachan, too, hopes for significant impetus from interdisciplinary cooperation: “Neuroscientists could use our neuromorphic systems to test their models of neural circuits and to better understand the brain,” he explains. Their ideas and experience, in turn, will help him and his team to further improve the hardware systems. In the end, each discipline enriches the others, and together they approach their great goal: a neuromorphic computer.
Janosch Deeg/Christian Hohlfeld
Photos: Forschungszentrum Jülich/Ralf-Uwe Limbach, Vorträge aus der Reihe „Wissenschaft online
