Artificial intelligence systems typically use lots of energy and rely on the cloud, which brings information security vulnerabilities.
Scientists in Switzerland have addressed these shortcomings by squeezing an AI system onto a single computer chip and supplying it with solar power.
Researchers at the Swiss Center for Electronics and Microtechnology, CSEM, presented their breakthrough AI system-on-chip at this week’s 2021 VLSI Circuits Symposium in Kyoto, Japan.
Both the ASIC chip and RISC-V processor used to make the new AI technology were developed by engineers at CSEM. The two components deploy novel signal processing architecture to minimize energy needed to power the AI algorithms.
Scientists programmed a pair of machine-learning accelerators onto the AI chip. One performs face detection, while the other handles classification.
The first accelerator uses what’s called a binary decision tree, or BDT, to carry out basic analysis tasks. However, the BDT engine is unable to execute recognition functions on its own.
When deployed for facial recognition, the BDT accelerator will pose and answer basic questions such as: Are people present in the images? If the answer is yes, the accelerator will prompt followup questions — questions farther down the tree branch — like: Are faces visible?
“If our system is used in voice recognition, the first accelerator will determine whether noise is present and if that noise corresponds to human voices,” Stéphane Emery said in a press release.
“But it can’t make out specific voices or words – that’s where the second accelerator comes in,” said Emery, head of system-on-chip research at CSEM.
The second accelerator, an engine known as a convolutional neural network, or CNN, is designed to assist on more complicated tasks, such as recognizing individual faces and discerning specific words.
The CNN engine uses a lot more power, but it only needs to run for short periods of time. By bifurcating — splitting — the two accelerators, researchers were able to drastically reduce the power requirements of the AI system.
Because the system is modular, it is highly adaptable.
“Our system works in basically the same way regardless of the application,” says Emery. “We just have to reconfigure the various layers of our CNN engine.”
The simple design, powered by either a battery or solar cell, can be integrated into a variety of environments and problem-solving scenarios. Scientists suggest the technology could be used to enhance everything from video surveillance systems to heart monitors.