International Business Machines Corporation
Thin-film large-area classifier

Abstract:

A classifier system implementing an equivalent deep neural network (DNN) includes a weight block, classification block, row selector, and sensor array coupled with the weight block, classification block and row selector. The sensor array includes row lines, column lines, a data integration line, an integration start line, and multiple sensor cells corresponding to respective neurons in an input layer of the equivalent DNN. The sensor cells share a common terminal connected to the data integration line, the row lines are controlled by the row selector, and the column lines receive respective weight values from the weight block. The classification block includes a first integrator receiving a signal generated on the data integration line when the integration start line is selected, and a first thresholding unit receiving a signal from the first integrator. The first thresholding unit is coupled to second integrators and second thresholding units arranged in a two-dimensional matrix.