back

»We want to rethink the way computer chips work, taking inspiration from natural processes in the brain.«

Prof. Dr.-Ing. Max Lemme conducts research on nanoscale transistors, 2D materials and perovskites for applications in communication, sensing and neuromorphic computing. Neuromorphic computing is the subject of the future cluster NeuroSys, which he is currently coordinating. NeuroSys stands for Neuromorphic Hardware for Autonomous Artificial Intelligence Systems and is one of seven future clusters selected for funding from over 130 submissions in the first round of the Clusters4Future initiative of the Federal Ministry of Education and Research (BMBF). The vision of the NeuroSys future cluster is to develop innovative hardware for applications in artificial intelligence (AI) and to establish the Aachen region as a global leader in this field.

Prof. Lemme, what does the NeuroSys Future Cluster stand for?

The NeuroSys Future Cluster has the long-term goal of leading basic research into application, as explicitly formulated in the Federal Ministry of Education and Research’s Cluster4Future funding line. We are tackling this task by bringing together the various competencies from RWTH Aachen University, scientific institutions, companies and public stakeholders from the region. We want to rethink the way computer chips work, taking inspiration from natural processes in the brain. This is of course a long-term goal, for which the cluster gives us a certain amount of freedom.

For this challenging task, we have covered the entire value chain in the region, from materials research, chip design, neuroscience and AI applications to ethical, sociological and business management issues. We have thus created various interfaces with regional companies along the chain and, despite the long timelines in research, can also realize transfer and innovation at short notice. We also see ourselves as part of the transformation process in the Rhenish mining area.

What makes neuromorphic hardware special and how does it differ from conventional hardware?

Neuromorphic computing is characterized by the fact that computer architecture and computing processes are inspired by the structure and functioning of the brain. A neuromorphic computer will therefore use artificial neurons to perform calculations. However, the term has since been expanded to include analog and digital systems (and combinations thereof) that implement models of neural systems.

Of course, one might now ask: why should we abandon the established methodology? The reason is that machine learning and artificial intelligence applications are extremely data intensive. In conventional computers, which are designed in the so-called » Von-Neumann-architecture «, the computing units and the memory are separated from each other, so that the huge amounts of data have to be constantly transported back and forth. This consumes an enormous amount of energy, which is exacerbated by the exploding number of users of these applications. The Semiconductor Research Corporation, an association of global semiconductor manufacturers and research institutes, predicts that we will spend all the energy generated on the planet on computing by around 2045 if nothing changes in computer hardware.

Talking about »energy efficiency«: Can you give us an example of where neuromorphic hardware can take AI applications to a new level?

In addition to the fundamental problem described above, there are a large number of application scenarios. Various »hardware accelerators« are already being used today, for example in smartphones, which are breaking new ground in design. One example for the future, however, is autonomous driving. To put it bluntly, vehicles would need as much energy to › think ‹ ‘ as they do to drive, which complicates the issue of range. Cloud solutions, on the other hand, are not yet feasible in real time and also pose a risk due to network reliability. Neurorphic chips could solve this. A second example is patient care, for example at the hospital bedside. Here, AI systems can provide massive support for medicine, but highly confidential data is collected that cannot be processed in a conventional cloud. But there is more to come in terms of networked industry, smart homes and smart cities: billions of sensors will collect data and, with the help of AI, lead to more productive, safer, more environmentally friendly processes, but the new hardware must also process these incredible amounts of data efficiently.

What societal challenges do you see in dealing with AI and how can neuromorphic software help to overcome these challenges?

Regionally, of course, the lignite phase-out and the necessary structural change in the Rhenish mining area is a challenge. We need new regional technologies that create sustainable jobs. This is a clearly defined goal of the cluster players, especially the companies and start-ups, of which we are pleased to say we already have a number on board. But of course, we are also thinking about technological sovereignty in an uncertain geopolitical situation in the European context. In my opinion, AI will have a significant impact on all areas of life, which is also a key finding of a Bundestag study commission on the subject. We therefore need to ask ourselves the question: how can we create an AI that benefits citizens, makes Europe more independent and generates decisions that are in line with our European values? The Human Technology Center at RWTH Aachen University deals with these ethical and sociological aspects of our research and is an elementary component of the cluster.

How does the work in the cluster look?

This brings me back to the value chain. The overarching goal of creating energy-efficient hardware for AI applications is a very long-term one. To this end, we have players from all parts of the value chain who are striving to co-design hardware software and algorithms. For example, we are researching completely new materials, but we are using them directly to create new electronic components as demonstrators. We use this measurement data to develop models for the functionality, which in turn can be used by colleagues in chip design to simulate future systems now. For neuromorphic computing on the virtual hardware, the software also has to be rethought. We are therefore trying to shorten the long timeframes through cooperation.

But the cluster also has a shorter-term dimension: innovations are constantly emerging at one level of the value chain, and our regional industrial partners and the supra-regional advisory board are on hand to turn research into products and ultimately jobs.

Who are you working with in the cluster?

My own work is in the field of electronic components made of so-called two-dimensional (2D) materials, for the discovery of which the Nobel Prize in Physics was awarded to Professors Geim and Novoselov in 2010. In the »Aachen Graphene & 2D Materials Center«, various chairs of electrical engineering and physics are researching these materials together with my »Institute AMO« , but the company Aixtron and the start-up Black Semiconductor are also very active here. Aixtron, as a manufacturer of process systems for semiconductor technology, can therefore further develop its technology portfolio, while Black Semiconductor wants to use this class of materials for its photonic chips. At the same time, we need new measuring methods for the characterization of new components, which the measuring equipment companies »AIXACCT« and »Amotronics« are developing with us and which they can then offer to their global customers in the form of their measuring equipment. These are very specific goals that benefit directly from the long-term research program. In addition, we can train the relevant experts through master’s theses and doctorates at RWTH in the cluster.

What role does networking in the region play for you?

Networking in the region is essential. I hope I was able to make it clear what a long-term, but also major goal we have in mind. Measured against this, the funding from the cluster is realistically a drop in the ocean, despite all modesty. We are talking about an amount in the single-digit millions per year, whereas the establishment of semiconductor companies requires billions of euros in support. We are therefore happy about any synergetic networking, for example with the structural aid project »NEUROTEC 2«, which is coordinated by Prof. Waser at the Research Center Jülich. In addition to companies, our advisory board also includes networking players such as KI.NRW, Zukunftsagentur Rheinisches Revier, AGIT GmbH, IHK Aachen and the City of Aachen. Exciting discussions and approaches on how the cluster can be further expanded are always taking place here.

Finally, a look into the future. Where is NeuroSys headed and what are your hopes for the coming years?

We are currently applying for the second of three phases of the cluster, in which we were able to increase the industrial share. I am particularly pleased that there are several start-ups among them that have already been able to raise venture capital, which in total significantly exceeds the BMBF funding. There are also start-ups from within the cluster, as well as an investment of up to 100 million euros in an R&D center by the company Aixtron in the region. In addition, there is a growing interest in the topic from large supra-regional companies, some of which are partners, some of which are supporters of our Industrial Graduate School and some of which are new members of the advisory board for phase 2. We are happy for this to continue. At some point, of course, we will need large semiconductor manufacturers to produce our neuromorphic chips. I hope that this will also happen in the Rhineland region. But I am also realistic and know that we can only achieve this with the support of the state and federal governments.