UW-Madison researchers created a sensor that processes early neural network layers inside each pixel using a P2M (processing-in-pixel-in-memory) design. This cuts energy use by 7.8× and memory by 25×, enabling efficient AI in small devices.
Month: June 2025
First Mamba Chiplet in GF 22nm
Engineers at UW-Madison have developed a cutting-edge Mamba chiplet using GlobalFoundries’ 22nm process. This breakthrough enhances hardware acceleration for state space models, significantly boosting energy efficiency and performance in sequential data tasks.