A modular programmable CMOS analog fuzzy controller chip

Abstract

Abstract-We present a highly modular fuzzy inference analog CMOS chip architecture with on-chip digital programmability. This chip consists of the interconnection of parameterized instances of two different kind of blocks, namely label blocks and rule blocks. The architecture realizes a lattice partition of the universe of discourse, which at the hardware level means that the fuzzy labels associated to every input (realized by the label blocks) are shared among the rule blocks. This reduces the area and power consumption and is the key point for chip modularity. The proposed architecture is demonstrated through a 16-rule twoinput CMOS 1-m prototype which features an operation speed of 2.5 Mflips (2.5 2 2 2 10 6 fuzzy inferences per second) with 8.6 mW power consumption. Core area occupation of this prototype is of only 1.6 mm 2 including the digital control and memory circuitry used for programmability. Because of the architecture modularity the number of inputs and rules can be increased with any hardly design effort.

Neuro-fuzzy chip to handle complex tasks with analog performance

“…Note also that this increased number of rules is not accompanied by a significant power consumption increase; neither by an operation speed drop. Actually, regarding power consumption, only the I m = 800mA prototype in [14] consumes less power than the MFCON chip, although it realizes four times less rules. Regarding speed, the prototype in [13] is faster, although its power consumption is much larger and its complexity much smaller.…”

“…For instance, a fully analog controller designed by the authors [14] using similar circuitry yields 470ns and 8.6mW for 16 rules, while the MFCON chip yields 500ns and 16mW for 64 rules. Advantages of the proposed architecture become more evident as the number of rules and inputs increases [1].…”

“…The circuitry is based on that previously reported by the authors in [14], although some important modifications have been made to incorporate programmability, as well as to save area and power. …”

“…On the other hand, they need the outer shell of analog circuitry to build the interface with sensors and actuators. Analog circuits [12][13] [14][15] [16] are considered good candidates to implement neural networks, despite their sensitivity to errors and noise, because the precision requirements are supposed low. However, the latter is not as true in ANN as in other networks with a high redundancy as the multilayer perceptrons, due to the fact that just a few nodes and parameters determine the output, thus errors in these nodes modify the output and are not compensated or corrected by other nodes.…”

“…Thus, if the input dimension grows and hence the system complexity, errors are difficult to keep bounded. The most complex pure monolithic analog fuzzy controllers implement around 15 rules (basis functions) [13] [14]. Nevertheless, since analog circuits provide the best efficiency in terms of area and power/speed ratio, it would be desirable to be able to increase their complexity to manage problems as motor control, where complexities above 25 rules are common [5].…”