Analogue fuzzy processor improves smart sensors

Analogue fuzzy processor improves smart sensorsSteve Bush Smart sensors have in-built circuitry that removes the need for external signal processing. For instance, a smart sensor may measure the vibration, temperature and current consumption of an electric motor and produce a signal which estimates how much wear and tear has occurred. As yet, few if any smart sensors exist in monolithic form. The sensor part is not the main problem. Silicon sensor structures are available for physical quantities including: acceleration, angular acceleration, pressure, force, temperature, light intensity and magnetic field strength – although in some cases the technology is not yet mature. It is the processing part of smart sensors which is still under debate. Fuzzy logic is an obvious choice for any sensor that has a number of input variables and has to combine them in some way, but how should the fuzzy processing be implemented? The processor has to be small, to keep total sensor cost low; dissipate very little power to avoid creating error inducing on-chip thermal gradients; and be compatible with the silicon process, mostly the CMOS process, used to make sensors. Dr Gillian Marshall of the UK’s DERA research laboratory thinks analogue fuzzy processors using flash-memory-like floating gate transistors are the answer. She cites an example of a two-sensor seven-rule digital fuzzy controller developed to balance an inverted pendulum. “This used 5,000 gates on an Asic which occupied over 4mm2 on a 1?m CMOSprocess,” said Marshall. Using DERA’s analogue process she claims that this could be reduced to 0.3mm2 using its proposed 2?m transistors. Fuzzy logic using low power analogue CMOS A team at DERA has developed a range of low power analogue CMOScircuits that can perform all the processing functions required in a fuzzy logic processor. The diagram shows two ways of making an ‘input fuzzifier’, the cell that implements the classic triangular transfer function that decides how much of a certain fuzzy set a particular input value belongs to. Programmability, which sets the way an input fits into a set, comes from using floating gate (shown in purple) transistors similar to those in flash memory – and programmed the same way. The transistors are operated in the sub-threshold region, also know as in weak inversion. “A digital circuit designer would describe them as off,” said Marshall.
The left hand circuit is an earlier design that has the disadvantage of a variable current consumption, between 0.1 and 10nA. The right hand cell uses grounded floating gate substrates to fixed consumption at 1nA and consume less average power.

Leave a Reply

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