1.2 nJ/classification 2.4 mm² asynchronous wired-logic DNN processor using synthesized nonlinear function blocks in 0.18 μm CMOS

Abstract We have developed a full-digital wired-logic DNN processor that is 5.3 times smaller and 2.6 more energy efficient than previously processors. Our capable of inferring MNIST classification task with 90.6% accuracy 1.2 nJ per at 3.89 Mfps. also neuron synapse-saving neural network using nonlinear technology to reduce the number processing elements be implemented. Lastly, we logical compression technique for area energy-saving cell circuits. Using these techniques, devised digital asynchronous processor.