Asynchronous or Analogue Methods for Computation
A talk from EMF 2016 by Matthew Ireland
On Sunday August 7, 2016 at – in Stage C
Every modern computer processes digital data on successive ticks of a synchronising "clock". We are all familiar with using the speed of this clock to reason about the performance of our devices: in 2000, a desktop might have run at 500MHz; now, it would not be unheard of to see a 5GHz clock. The existence of this clock makes processor design very convenient, and in the early days clocked designs generally proved faster and used fewer components than their unclocked counterparts.
However, in recent years we have seen a plateau in increasing clock speeds: computers are no longer getting "faster", per se. We have uncovered many disadvantages of using clocked circuits: the clock is power-hungry, it is difficult to distribute across a modern chip, all circuits are constrained to operate at the speed of the slowest, and clocks may cause electromagnetic interference to other nearby devices.
Taking inspiration from the human brain, this talk will examine two fundamentally different methods of computation that are either asynchronous or analogue in nature. They do not rely on a clock to advance the state. We shall simulate and explore these methods through interactive, graphical means. Asynchronous circuits have the potential to be faster than clocked designs and require far less power. Electronic analogue computers have the further advantage that they do not suffer from numerical errors that arise from approximating real numbers using a finite number of bits. Might one of these techniques shape the future of our technological society?