299 OLAF-8: Bounded-Memory Online Adaptive Fuzzy Inference :: Quicker, easier and cheaper to make your own chip!

Credits

We gratefully acknowledge the Center of Excellence (CoE) in Integrated Circuits and Systems (ICAS) and the Department of Electronics and Communication Engineering (ECE) for providing the necessary resources and guidance. Special thanks to Dr. H V Ravish Aradhya (HoD- ECE), Dr. K R Usha Rani (Associate Dean-PG), Dr. K. S. Geetha (Vice Principal) and Dr. K. N. Subramanya (Principal) for their constant encouragement and support in facilitating this Tiny Tapeout SKY26C submission.

OLAF-8: Bounded-Memory Online Adaptive Fuzzy Inference

Project Overview

OLAF-8 is a compact fully digital fuzzy inference engine designed to demonstrate online-adaptive fuzzy reasoning within a strictly bounded hardware memory.

The architecture processes two 4-bit input values and evaluates them using fuzzy membership functions and an eight-entry rule memory. Unlike a conventional fixed-rule fuzzy inference system, OLAF-8 can dynamically modify its rule memory during operation. When the current input is insufficiently represented by the stored rules, the architecture identifies a low-utility rule and replaces it with a rule derived from the current input.

The design combines:

4-bit dual-input fuzzy processing
LOW, MID and HIGH membership regions
Eight-entry bounded fuzzy rule memory
Sequential rule evaluation
MIN-based rule firing strength
Weighted fuzzy accumulation
Iterative digital division for defuzzification
Maximum firing-strength tracking
Utility-based rule selection
Online rule admission and replacement

The key design concept is bounded online adaptation: the system can update its fuzzy knowledge during operation while maintaining a fixed eight-rule hardware memory. This avoids the need for an external processor or complete hardware reconfiguration for adapting the rule set.

OLAF-8 is implemented entirely in the digital domain and is intended as a compact research demonstrator for exploring adaptive fuzzy inference under strict hardware resource constraints suitable for Tiny Tapeout.

How it works

OLAF-8 is a compact fully digital fuzzy inference engine with an 8-rule bounded rule memory and online rule admission.

The design accepts two 4-bit input values, x1 and x2.

For each input transaction:

The input values are captured.
Membership degrees for LOW, MID and HIGH fuzzy regions are calculated.
The eight stored fuzzy rules are evaluated sequentially.
Each rule firing strength is calculated using the MIN operation.
The weighted fuzzy output is accumulated.
The maximum rule firing strength is tracked.
An iterative divider calculates the defuzzified output.
If the maximum firing strength is below the admission threshold, the least-utility rule is replaced with a rule derived from the current input.
Otherwise, the winning rule utility is increased.

The rule memory is permanently bounded to eight entries. This allows online adaptation without requiring an external processor or reconfiguration.

The design is intended as a fully digital Tiny Tapeout research demonstrator.

How to test

The OLAF-8 interface uses the standard Tiny Tapeout pins.

Inputs

ui_in[7:4] - 4-bit input x1
ui_in[3:0] - 4-bit input x2
uio_in[0] - Start signal

Outputs

uo_out[3:0] - 4-bit fuzzy output
uo_out[4] - Done pulse
uo_out[5] - Rule admission/replacement indicator
uo_out[6] - Busy indicator
uo_out[7] - Reserved

To execute a transaction:

Wait until busy is low.
Apply the desired 4-bit x1 and x2 values.
Generate a one-clock pulse on uio_in[0].
Wait until uo_out[4] becomes high.
Read the fuzzy output from uo_out[3:0].
Check uo_out[5] to determine whether online rule admission occurred.

The design uses sequential rule evaluation and iterative arithmetic, so the output is generated after multiple clock cycles rather than immediately.

External hardware

No external hardware is required.

OLAF-8 is a fully digital design and is intended to operate using the standard Tiny Tapeout clock, reset, input and output interfaces.

#	Input	Output	Bidirectional
0	x2[0]	fuzzy_y[0]	start
1	x2[1]	fuzzy_y[1]
2	x2[2]	fuzzy_y[2]
3	x2[3]	fuzzy_y[3]
4	x1[0]	done
5	x1[1]	rule_admitted
6	x1[2]	busy
7	x1[3]	reserved

299 OLAF-8: Bounded-Memory Online Adaptive Fuzzy Inference