Two AI processors for voice assistants and medical devices turned out to be 10% more powerful and 90% more energy efficient than last year’s analogues.
According to SCMP reports, scientists from China University of Electronic Science and Technology presented two artificial intelligence (AI) chips with ultra-low power consumption and record performance.
Chips designed specifically to perform AI tasks often require significant amounts of computing power. In this respect, they spend a lot of energy. But through optimization of algorithms and architecture, Professor Zhou Jun from the University of Electronic Science and Technology of China and his team were able to significantly reduce processor power consumption.
Researchers presented two innovative chips at the IEEE International Solid State Circuits Conference ISSCC. Every year, the event brings together the best researchers, engineers and experts to discuss new developments and the future of chip technology. This year the conference was held in San Francisco (USA) from 18 to 22 February.
The first of two AI chips is designed to be built into smart devices and provide voice control while offline. Thanks to this, the device does a good job of detecting keywords and controlling live speech, recognizing the speaker’s voice signals even in the presence of external noise. Zhou’s team proposed a new architecture that includes dynamic computing engines, an adaptive noise reduction circuit, and an integrated keyword and speaker recognition circuit.
The chip provides a recognition power consumption of less than two microjoules per sample, with accuracy exceeding 95% in quiet environments and 90% in noisy environments. The processor will be useful for low-power voice control applications such as smart home, wearable devices and smart toys.
The second chip presented at the conference was designed to detect epileptic seizure signals. Designed for use in wearable devices, the device uses electroencephalogram (EEG) recognition to alert the patient to an epileptic seizure so they can seek medical attention.
Researchers achieved over 98% accuracy by optimizing a retraining algorithm that allows a pre-trained AI model to make accurate predictions based on existing data without the need to collect patient seizure signals. Before use, patients simply wear the device and perform calibration; This process only takes a few minutes. This allows the device to recognize individual signal characteristics. The average power consumption of this chip during recognition is only around 0.07 microjoules, making it the most energy-efficient configuration among such processors.
In the official report, the scientists stated that their chip is 10% more powerful in terms of computing and 90% more efficient in terms of energy than a similar processor presented at last year’s conference.
Important
During the demonstration at ISSCC, EEG signals collected in real time from the user’s wearable brain-computer interface device were transmitted via Bluetooth to a test board. The chip was reconfigured to identify imaginary motor commands, allowing the robot to be controlled to move forward, stop or go backwards.
We previously wrote that the USA would punish China for circumventing sanctions and Huawei smartphones would be deprived of advanced chips. The White House plans to tighten restrictions on SMIC and Huawei by involving Germany and South Korea in the sanctions war.
Source: Focus
Ashley Fitzgerald is an accomplished journalist in the field of technology. She currently works as a writer at 24 news breaker. With a deep understanding of the latest technology developments, Ashley’s writing provides readers with insightful analysis and unique perspectives on the industry.