Laboratoires d’Electronique Philips has come up with a digital VLSI circuit called L-neuro which features an on-chip learning capability and can be used to implement neural networks. According to Electronic World News, the chip learns the relevant features depicted in an image in real time – the knowledge can then be passed on to a principal component analysis algorithm for image compression. L-neuro is fabricated in 1-micron CMOS and can store 1,024 neuron connections in random access memory. A typical response time for updating an 8-bit neuron is one nanosecond, irrespective of the size of the network. Although Philips says that the chip is part of a research project and that there are no plans to manufacture or market the device, the L-neuro chip’s learning capacity puts it well ahead of Intel’s N64 neural chip which has yet to see the light of day. For the N64, shipment of which has been delayed by bugs (CI No 1,392), needs supporting software in order to train it in the patterns of the type of data for which the neural network will be used. Philips says it has been focussing strongly on neural network applications in its work on image compression for television.
