China makes super energy-efficient AI chips

Two new AI chips enable offline voice control and seizure detection with very low power consumption.

AI chips used for heavy tasks often require large amounts of power, limiting their practicality. Professor Zhou Jun and his colleagues at the University of Electronic Science and Technology of China (UETC) developed two new types of AI chips and introduced that they are the most energy efficient in the world, Interesting Engineering on March 25 reported.

A small computer chip. (Photo: Gulnara Mandrykina/iStock).

The team significantly reduced their power consumption through algorithmic and structural optimization. Two new AI chips were introduced during the IEEE 2024 International Solid-State Circuits Conference (ISSCC) taking place in San Francisco, USA.

The first AI chip is designed to be placed in smart devices, making offline voice control possible. According to the research team, the chip is excellent at detecting keywords and verifying speakers by recognizing the target speaker's voice signal. Its main advantage is its ability to overcome the limitations of standard speech recognition systems. It can accurately recognize the target speaker's words even in noisy situations with distractions such as TV, music or chat.

"This chip achieves recognition power consumption of less than 2 microjoules at a time, with an accuracy rate of more than 95% in quiet and 90% in noisy environments, helping to set new global standards. in terms of both energy efficiency and accuracy ," the research team explains.

In a test demonstration, a 1cm ² chip was integrated into the microcontroller inside the toy car to control its movement. The chip can also be used for low-power voice control scenarios such as smart homes, wearables and smart toys.

The research team's second AI chip helps detect seizure signals in epilepsy patients. Designed for wearable devices, it uses EEG recognition to detect seizures and alerts, so patients can seek medical assistance. With an average recognition power consumption of just 0.07 microjoules, it is the world's most energy-efficient AI chip of its kind.

Compared to another chip introduced at last year's ISSCC conference, the new chip increases accuracy by 10% and reduces power consumption by more than 90%. In addition to seizure detection, it has other potential applications such as brain-computer interfaces and sleep monitoring.