AI is used in a variety of very useful applications, such as predicting machine life through vibration, monitoring patient cardiac activity, and incorporating facial recognition into video surveillance systems. The downside is that AI-based technologies generally require a lot of power and, in most cases, need to be permanently connected to the cloud, creating issues related to data protection, IT security, and energy use. ..
CSEM engineers may have found a way around these issues, thanks to the new system-on-chip they have developed. It runs on a small battery or a small solar cell and performs AI operations on the edge. That is, it runs locally on the chip, not in the cloud. In addition, these systems are fully modular and can be tailored to any application that requires real-time signal and image processing, especially when sensitive data is involved. Engineers will be presenting their devices at the prestigious 2021 VLSI Circuit Symposium in Kyoto this June.
CSEM engineers have developed integrated circuits that can perform complex artificial intelligence operations such as face, voice, gesture recognition, and heart monitoring. Equipped with either a small battery or a solar panel, it can be configured to process data at the edge and be used in almost all types of applications. Credit: CSEM
The CSEM system-on-chip works through a brand new signal processing architecture that minimizes the amount of power required. It consists of a RISC-V processor (also developed in CSEM) and an ASIC chip with two tightly coupled machine learning accelerators. One is for face detection and the other is for classification. The first is the Binary Decision Tree (BDT) engine, which can perform simple tasks but not recognize operations.
“For example, if our system is used in a facial recognition application, the first accelerator will answer preliminary questions such as: Are there people in the image? If so, their faces are visible. “?” Says Stéphane Emery, head of system-on-chip research at CSEM. “When our system is used in speech recognition, the first accelerator determines if there is noise, and if that noise corresponds to the human voice, but a particular voice. I can’t understand the words and words, so a second accelerator comes in. “
The second accelerator is the Convolutional Neural Network (CNN) engine, which can perform these more complex tasks (recognition of individual faces and detection of specific words), but consumes more energy. In most cases, this two-tier data processing approach significantly reduces the power requirements of the system because only the first accelerator is running.
As part of their research, engineers have improved the performance of the accelerator itself, making it adaptable to any application that requires time-based signal and image processing. “Our system works basically the same regardless of the application,” says Emery. “We need to reconfigure the various layers of the CNN engine.”
CSEM’s innovation opens the door to a whole new generation of devices with processors that can run independently for over a year. It also significantly reduces the installation and maintenance costs of such devices, allowing them to be used in locations where battery replacement is difficult.
New energy-efficient AI system on-chip powered by solar power
https://scitechdaily.com/new-energy-efficient-ai-system-on-chip-runs-on-solar-power/ New energy-efficient AI system on-chip powered by solar power