LightTok 2D chip converts light directly into AI tokens with 10x efficiency

Researchers at Nanjing University unveiled LightTok, a 2D chip that converts raw light directly into tokens for on-chip AI models, skipping five energy-intensive processing stages. It reached 87.3% image recognition accuracy and is 10 times more energy efficient, though resolution is limited to 32x32 pixels.

• Chip merges sensing, memory and computation into a single pixel using molybdenum disulfide
• Image recognition accuracy reached 87.3%
• Light-to-token conversion is 10 times more energy efficient
• Current resolution is 32x32 pixels, with CMOS scaling seen as possible