Artificial intelligence systems typically use a lot of energy and rely on the cloud, leading to information security vulnerabilities. Swiss scientists have remedied these shortcomings by installing an artificial intelligence system on a single computer chip and supplying it with solar energy.
Researchers at the Swiss Center for Electronics and Microtechnology (CSEM) demonstrated at the 2021 VLSI Circuits Symposium in Kyoto their breakthrough AI system-on-chip system, UPI reported. The ASIC chip used for the new AI technology and the RISC-V processor were also developed by CSEM engineers. The two components use a new signal processing architecture to minimize the energy required to operate AI algorithms.
Scientists have programmed a pair of machine learning accelerators on the AI chip. One performs facial recognition while the other performs grading. The first accelerator uses a so-called binary decision tree (BDT) to perform basic analysis tasks. However, the BDT engine alone cannot perform recognition functions. The BDT accelerator used for face recognition asks and answers basic questions such as: Are people present in the images? If the answer is yes, the accelerator asks additional questions – questions below the tree branch – such as: Are faces visible?
“If our system is used for voice recognition, the first accelerator determines whether this noise, and whether this noise corresponds to human voices. But you can’t distinguish specific sounds or words. Here comes the second accelerator, “said Stéphane Emery, CSEM’s research leader.
The second accelerator, an engine called a convolutional neural network (CNN) was designed to help with more complex tasks, such as recognizing individual faces and distinguishing specific words. The CNN engine uses a lot more power, but it only needs to run for a short time. By separating the two accelerators, the researchers were able to drastically reduce the energy requirements of the artificial intelligence system. Because the system is modular, it is extremely adaptable.
“Our system basically always works the same, regardless of the application. We just need to reconfigure the different layers of our CNN engine,” said Emery. The simple, battery- or solar-powered design can be integrated into a variety of environments and problem-solving scenarios. Scientists say the technology could be used for everything from video surveillance systems to heart monitors.
Hardware, software, tests, curiosities and colorful news from the world of IT by clicking here