Scientific Achievement
- Researchers in the Electronic Materials program determined novel photodetectors can learn and perform machine learning spectral analysis inside the sensor
Significance and Impact
- Spectral machine vision inference directly via analog photocurrent eliminates post-processing, reducing power consumption and latency by more than two orders of magnitude
Research Details
- Developed mathematical theory and learning algorithm for in-sensor spectral kernel computation
- Black phosphorus-MoS2 photodiode allows enables visible-to-MIR spectral machine learning for diverse tasks, including image segmentation, natural/artificial leaf identification, and chemometric analysis
Publication Details
D. Zhang, Y. Li, J. Geng, H. M. Kim, M. Ma, S. Wang, I. Kim, T. J. Wijaya, N. Higashitarumizu, I. K M R Rahman, D. Urmossy, J. Bullock, A. Ozcan, A. Javey, Science (2025).
Work was performed at Lawrence Berkeley National Lab.
