China Unveils ACCEL: An Analogue AI Chip Outperforming Nvidia's A100 in Vision Tasks

Breakthrough in Analogue AI Computing

Researchers from Tsinghua University in China have announced the development of a novel analogue artificial intelligence chip named ACCEL (All-analog Chip Combining Electronic and Light Computing). This innovative photoelectronic chip is designed to significantly enhance the speed and energy efficiency of computer vision tasks, potentially surpassing the capabilities of current high-performance GPUs like Nvidia's A100. The findings of this research were published in the scientific journal Nature.

Performance and Efficiency Benchmarks

In laboratory tests focused on computer vision tasks, ACCEL demonstrated remarkable performance metrics. The chip was found to be up to 3,000 times faster than Nvidia's A100 GPU, achieving a computing speed of 4.6 peta-floating point operations per second (PFLOPS). Furthermore, ACCEL exhibited extraordinary energy efficiency, consuming 4 million times less energy than a top-of-the-line GPU for similar tasks. Its accuracy in recognizing and classifying objects was comparable to that of digital neural networks.

Innovative Architecture and Fabrication

ACCEL's superior performance stems from its unique architecture, which integrates both photonic (light-based) and electronic computing in an all-analogue manner. This design avoids the need for analog-to-digital conversion, a process that is typically time- and energy-intensive in conventional digital chips. According to Fang Lu, a researcher from the Tsinghua team, 'We maximized the advantages of light and electricity under all-analog signals, avoiding the drawbacks of analog-to-digital conversion and breaking the bottleneck of power consumption and speed.'

The chip's development also highlights China's strategic advancements in semiconductor technology. ACCEL was fabricated using a 20-year-old transistor fabrication process by China's Semiconductor Manufacturing International Corporation (SMIC). This approach reduces reliance on advanced lithography machines, which are often subject to international restrictions.

Potential Applications and Future Outlook

While ACCEL is specialized for specific tasks and cannot perform general computing functions like those found in smartphones, its capabilities open doors for numerous applications. Potential areas include:

• Wearable devices
• Electric cars
• Smart factories
• Unmanned systems
• Industrial inspections
• Large-scale artificial intelligence models

Dai Qionghai, Director of the School of Information Science and Technology at Tsinghua University, emphasized the importance of translating this new architecture into practical applications to address 'major national and public needs.' The development of ACCEL positions China to enhance its competitive edge in the mass application of AI, particularly in areas like image recognition and autonomous driving.