An Electronics Weekly article gives an overview of two current projects to implement hardware-based neural network processing. One is a shipping product, the other is still in research and development stages. Doing neural networks in hardware is much faster than implementing them in software. Mentioned are Axeon's Vindax VX1064 ANN coprocessor which provides 64 physical neurons in an SIMD array. The processor can be used to implement neural networks of arbitrary size by mapping them onto the physical neurons. It has achieved speeds of 500k classifications per second. They also describe a project by Steve Furber, one of the principal architects of the ARM processor. Furber's project is developing a chip called FIRE. Each of the chips would contain a PGA and multiple ARM CPUs and might be capable of providing thousands of neurons. They hope to use the chips as nodes in a larger processor to achieve 1 million hardware neurons.