Engineers at the Swiss Center for Electronics and Micro technology (CSEM) have developed a system-on-a-chip that performs AI operations locally and can run on a small battery or solar cell.
While AI in its wildly varying forms is becoming ubiquitous, AI-based technology has the drawbacks of generally requiring a lot of power to operate and being permanently connected to the cloud, increasing data protection, energy efficiency and security concerns.
The team of CSEM engineers has taken a step towards solving these problems by developing a system-on-a-chip that consumes relatively little power and performs AI operations at the edge (on-chip rather than from in the cloud).
The system uses an all-new signal processing architecture that minimizes power consumption. It consists of an ASIC chip with a RISC-V processor (another CSEM innovation) and two closely linked accelerators for machine learning:
one for face detection, for example, and the other for classification. The first is a binary decision tree engine for performing simple tasks, but not recognition operations.
“For example, if our system is used in facial recognition applications, the first accelerator answers preliminary questions like, ‘Are there people in the images? And if so, are their faces visible?” says Stéphane Emery, head of system-on-chip research at CSEM.
“When our system is used in speech recognition, the first accelerator determines whether noise is present and whether that noise matches human voices, but it can’t distinguish specific voices or words, that’s where the second accelerator comes in.
” The second machine learning accelerator is a convolutional neural network engine for performing more complex tasks, such as recognizing individual faces or specific words. This accelerator – which consumes much more energy – turns only when needed.
This two-pronged data processing approach reduces the average power consumption of the system, allowing it to operate on a small battery or solar cell.As part of their study, the engineers then improved the performance of the accelerators themselves.
They made the system very versatile; it is modular, which allows it to be adapted to any application that requires real-time signal and image processing.”Our system basically works the same way regardless of the application,” Emery says.
“We just need to reconfigure the different layers of our CNN engine.” a-chip architecture could open the door to a whole new generation of devices with t processors that run independently for more than a year.
It could also reduce installation and maintenance costs for this class of devices and use them in locations where it would be impractical to change the battery regularly.