Looking for a new computer

In search of a new computer architecture

By JANO COSTARD, DR. JULIA SCHNEIDER, illustrations by NOËLLE KRÖGER

February 22, 2023 · Why the typical universal computer is increasingly reaching its limits – and what an innovation competition can bring.

Swarms of artificial brains that support everyday life and the world of work will be woven into the digital world of the future. Data from countless sensors and actuators are read out by the smallest computers, processed, brought together via a network of data paths in larger nodes with more computing power, interpreted – and returned. In between are huge data centers that take over the big tasks, manage and direct data, train artificial brains and solve complex scientific problems.

But this hope for a human-sustaining, networked world currently comes at a high price. If energy consumption for computing and communications continues to grow at the same rate as before, it will take up the entire capacity for energy production by 2040.

For more than five decades, the so-called Moore's Law applied in chip production. This states that due to rapid development of the technology, the number of transistors on an area doubles approximately every 18 months. Processors became faster and smaller due to the technically demanding reduction of the structures on the chip. Computing power became cheaper. Inspired by Moore's Law, the chip industry followed a coordinated roadmap for decades, the International Technology Roadmap for Semiconductors (ITRS), which defined the next development steps for all sub-areas, so that all stakeholders were prepared to make immense investments. On this basis, this unique industrial sector drove the progress of information technology.

There is a lack of impulse and wake-up call

The most widespread calculation principle is the so-called von Neumann architecture, whose central characteristic is that there is a common memory for instructions and data. Data and programs are accessed sequentially through a single channel. However, the common channel, for whose bandwidth commands and data compete, creates a bottleneck for the performance of the system. The typical universal computer, on the desk or in the trouser pocket, uses the von Neumann architecture. A solution to this problem through incremental hardware development is not foreseeable. The "von Neumann bottleneck" can therefore apparently only be eliminated by fundamentally new computing concepts.

In addition to quantum computing, there are alternatives to the von Neumann architecture with new ideas for neuromorphic computing, analog computing and a number of other unconventional approaches. However, these approaches are mostly in the research stage and – with the exception of quantum computing – live in a niche away from large funding programs or investments by industry. In view of the major technological challenges and the associated uncertainties in the development of these new technologies, the necessary investments are not only associated with high financial risk. They would also have to assert themselves against investment decisions along the innovation path that has been followed for decades. Thus, the previous strength of the semiconductor industry, the aforementioned coordination of investments via a common roadmap, is now becoming a major hurdle for change.

What is missing, therefore, is not only the financing of research and development of new computing concepts. It also lacks an impulse and wake-up call to activate and bring together those who can make new computing concepts a reality. The Federal Agency for Leapfrog Innovations SPRIND now wants to fill this gap with an "SPRIND Challenge" for the development and implementation of new computing concepts. In the first phase of this international innovation competition, 14 teams will develop new computing concepts, which will be implemented from mid-2023 in the subsequent second phase using appropriate hardware.

Taking high risks

Experience has shown that innovation competitions can be condensation nuclei for the development of new technologies and industries. A prototypical example of innovation competitions are the Grand Challenges of the American Defense Advanced Research Projects Agency (Darpa). These innovation competitions from the 2000s have led to a technological leap in the field of autonomous driving that would hardly have been possible without these DARPA challenges. On this basis, not only have new players such as Google's subsidiary Waymo emerged, but technologies around autonomous systems, such as lidar, have also fundamentally evolved.

At that time, the impetus for profound technological change was not primarily taken up by the established German car manufacturers, but by American technology companies such as Google. The fact that the automotive industry has not yet been disrupted by autonomous systems from American technology companies only shows how high the technological hurdles are for these leapfrog innovations.

The Federal Government's mandate to the Federal Agency SPRIND to take correspondingly high risks in the development of groundbreaking new technologies. We are sure that we have the expertise to develop new computing concepts from Germany and Europe and to build up a strong international competitive position. The SPRIND Challenge "New Computing Concepts" is intended to initiate this development.

Dr. Jano Costard is Challenge Officer of the Federal Agency for Leapfrog Innovations SPRIND.

The comic was created in cooperation with the Federal Agency for Leapfrog Innovations SPRIND. Last year, it launched an innovation competition on new computer architectures.

Future of IT Human-compatible Metaverse Into the Matrix