ZHAW researchers develop AI platform for open processor architecture
ZHAW researchers are developing a software and hardware platform based on the RISC-V open source processor architecture as part of the EU-funded project REBECCA. The aim is to ultimately make the platform available for a wide range of applications, primarily in the area of the Internet of Things, and in doing so help to further expand the computer chip industry in Europe.
The manufacturers Intel, AMD and, in particular, ARM have been the dominant names on the computer chip market to date. ARM chips can be found in almost all smartphones and even the chips now developed by Apple itself are based on ARM architecture, for which expensive licences have to be acquired. However, ARM’s de facto monopoly could be toppled in the not too distant future by RISC-V, an open architecture for computer processors. The development of RISC-V (reduced instruction set computer) began at the University of California in Berkeley in 2010, and the architecture is licence-free, meaning it can be used at no cost. This also makes it an interesting proposition for major players like Google, which has now also ported its Android operating system to RISC-V in addition to ARM and x86. The EU-funded REBECCA project (Reconfigurable Heterogeneous Highly Parallel Processing Platform for safe and secure AI) is now aiming to give RISC-V an additional boost.
Promoting the chip industry in Europe
The ZHAW School of Engineering is represented as part of the EU-funded project by Matthias Rosenthal, Head of the Realtime Platforms Research Area at the Institute of Embedded Systems (InES), and Hans Dermot Doran, Head of the High Integrity Systems Research Group (also part of the InES). “One of the aims of REBECCA is to promote the RISC-V architecture through its work and ensure it receives greater attention, as Europe is by no means one of the pioneers in the area of computer chip production,” says Matthias Rosenthal. A total of 24 research and industrial partners wanting to industrialise the technology are involved in the project, including the Polytechnic University of Turin, the Fraunhofer Institute in Germany and the Turkish household appliance manufacturer Arçelik. The EU-funded project, which has a total budget of EUR 8.4 million, is scheduled to run for three and a half years until summer 2026.
Applications for drones and refrigerators
Within the project, applications for four use cases are being developed that cover a relatively broad range of uses. “It is primarily about sensors and actuators for the Internet of Things,” explains Hans Dermot Doran. Among other things, the ZHAW team is developing safety applications, so-called processing-on-the-edge applications for underwater drones that are to be used for port inspections. A further use case focuses on air-based drones that are earmarked for use as part of inspection flights at photovoltaic plants. For another use case, smart on-the-edge processors are being developed for refrigerators that use cameras to recognise which foods are currently inside them. On the edge means that the computing power is located directly in the processor and the data does not first have to be sent to an external server for processing. This is an advantage for real-time applications.
ZHAW developing AI algorithms and safety components
The ZHAW’s research work as part of REBECCA is divided into two development groups that are each funded with CHF 300,000. “One group is focussing on AI on the edge, i.e. optimising real-time AI applications for this platform,” explains Matthias Rosenthal. Hans Dermot Doran’s group, in contrast, is working on the area of functional safety. “We are responsible for the reliable and safe execution of algorithms,” says the ZHAW researcher. “In concrete terms, this means that our algorithms aren’t tasked with delivering the right results, but rather with providing reliable data that has been calculated correctly,” states Doran. “When using inspection drones, for example, our algorithms ensure reliable calculation processes, the results of which are not distorted by noise or other disturbances,” the Head of the High Integrity Systems Research Group goes on to explain. In the future, these algorithms will also be used in safety-critical systems, for instance. “The difficulty here is that our algorithms have to work efficiently despite unfavourable architectural conditions and ultimately deliver accurate calculations,” says Hans Doran. The tasks of Matthias Rosenthal’s group are no less challenging. “In light of the low capacity in terms of memory and energy in edge processors, we have to proceed extremely economically in optimising the resources of hungry real-time AI applications,” states Rosenthal. Especially in the case of sensors that work on a practically energy-autonomous basis for an extended period, energy and computing power have to be used sparingly, as the sensors and AI algorithms nevertheless have to work reliably at all times. “The task here is to strike a precise balance between functionality and the use of resources,” says Matthias Rosenthal aptly in summing up his research task.