Researchthermal computingmatrix multiplicationinverse designthermal management
MIT Demonstrates Heat-powered Matrix Multiplication Devices
8.2
Relevance Score
MIT researchers Caio Silva and Giuseppe Romano publish in Physical Review Applied that inverse-designed microscopic silicon structures can perform matrix-vector multiplication using heat rather than electricity, achieving over 99% accuracy in simulations. The pore-filled, dust-sized devices guide thermal flow to compute and could enable passive overheating detection and low-power thermal management, though bandwidth and scaling limit complex deep-learning use.
