Programming CNN: a hardware accelerator for simulation, learning, and real-time applications

The cellular neural network (CNN) is a framework for cellular analog multidimensional programmable processing arrays with distributed logic and memory. The programmable feature is studied and its emulation with a low-cost high-speed hardware accelerator is described. The accelerator board, implemented as a multiprocessor PC add-on-board, is capable of handling one million processing cells with a speed of one million iterations per cell per second. It is part of a CNN workstation serving as a development system for CNN algorithms. The typical use of the CNN workstation for simulation, learning, and real-time applications is presented. As a simulator, nonlinear templates can also be emulated, and a sequence of series and parallel templates can be applied. Two application examples are described: textile pattern failures and printed circuit board (PCB) layout errors as determined by CNN template sequences