Clever chip can do quick maths

Clever chip can do quick maths
Richard Ball A microprocessor that uses logarithms to double the speed of computations is being designed as part of a European Esprit project. Developed by researchers at Newcastle upon Tyne University, the physical design of the chip is being carried out by Philips Research in Eindhoven. Silicon and development boards are expected in two year’s time. Applications set to benefit from this approach include mobile phones, computer graphics and high definition TV. “The feasibility study has all been done and we’ve simulated it to death,” said Dr Nick Coleman, project leader from Newcastle upon Tyne University. “We’re now going to manufacture a microprocessor around the new arithmetic unit.” “The basic object is to double the rate at which computers process the basic addition, subtraction, multiply and divide operations,” said Coleman. In floating point maths these operations, especially divide, are slow. “But when you use a logarithmic approach, the logarithm itself is fixed point,” he explained. The advantage of using logarithms is that floating point multiplies and divides become fixed point additions and subtractions. The former is around five times faster, the latter 15 times quicker. Since addition and subtraction are no faster using logs, the overall execution time of typical algorithms is halved. Accuracy is also improved. Using floating point for Fourier transform calculations for example, the final result can be several bits out. In contrast, fixed point operations produce no error.


Leave a Reply

Your email address will not be published. Required fields are marked *

*