Does weight-sharing lookup-table inference beat an int8 MAC engine once small arrays cost less to read?

research/accel-lutsize

Pricing every array by its own size flips accel-lut's negative result once read energy grows faster than size^0.26 (dense) or size^0.28 (pruned); Horowitz's slope is 0.5 above 8 KB but there is no point below it, so the verdict is conditional

7 tests

What it found

accel-lut found a product-table engine loses to an int8 MAC engine (about 2.7× the energy within a 2-point accuracy drop) while charging every array a flat 0.16 pJ/bit. accel-horowitzcheck showed that flat read ignores array size. This project prices each array (weights or indices, activation buffer, the 32-byte product table) as (10 pJ/64) (size/8 KB)^α per bit, keeping accel-lut's accuracy and nonzero counts. The cheapest design within tolerance ties the MAC at α = 0.26 (dense) and 0.28 (75% pruned); at α = 0.5, the slope of Horowitz's Fig. 1.1.9 between 8 and 32 KB, it uses 0.52 and 0.68 of the MAC's energy. Every array here is below 8 KB, where the paper has no data, so α is swept rather than chosen and the flip is conditional; a model that never prices arrays below the 8 KB point reproduces the original negative result. Next step: measure small arrays in an open PDK.

A figure from its own output

Figure produced by research/accel-lutsize: ratio — accel-lutsize/experiments/ratio.png, written by the project's own experiment code.

Paper, code and results

More in Task-specific AI accelerators

Is the OpenRAM access energy in SKY130 liberty files an energy at all, and does accel-pdkprice's weight-stationary verdict survive?OpenRAM's clk0 figure is an average power in mW over one cycle of an unstated period, not pJ: the unpacked verdict (register-file WS wins) holds for any period above 3 ns, while the packed verdict flips between 1 and 31 ns, so accel-pdkprice's packed conclusion is conditional
Once the MFCC front end is counted, does the energy-optimal keyword model change?Joining accel-cascade's committed results with accel-frontend's 46.8 nJ per frame: cascades' classifier-only saving of up to 28% becomes at most 0.42% of whole-system energy, while the most accurate model (about 0.75 test accuracy) costs 23 to 26 times the cheapest classifier but only 5% more system energy (modelled, not measured)
How big a keyword-spotting model fits a real Tiny Tapeout tile once the hidden layer goes past 64 units?Synthesising the streaming keyword-spotting engine at 96, 128, 192 and 256 hidden units (SKY130 HD, Yosys): area grows about 250 µm² per unit, so H = 128 (0.752 accuracy, 4-bit) needs a 2x2 tile (75,603 µm²) against a 1x2 (36,347 µm²) for H = 64 (0.737): +0.015 accuracy for twice the tile; H = 256 needs 4x2 at 50% utilisation, and its 8-bit layer-2 accumulator overflows 20 bits on a random model.
Do trained keyword-spotting models fit a Tiny Tapeout tile's 16-bit bias and 20-bit accumulator?All 96 retrained keyword-spotting models fit the Tiny Tapeout engine's 16-bit bias and 20-bit accumulator limits (worst use 43% and 30%), accuracy through the tile contract equals the reference, and 32 trained models run bit-exact in RTL.
Does Tiny Tapeout's real tile-size table change which keyword-spotting model fits a tile?Tiny Tapeout's real SKY130 tile sizes (1x1 is 161 × 111.52 µm = 17,955 µm², 0.43% below the value assumed earlier) applied to the synthesised keyword-spotting engine and the retrained models: the best 1x1 model is unchanged at 100% and 50% utilisation and flips at 70% (H = 16, 4-bit, 0.596 instead of 8-bit, 0.616, a 29 µm² miss); H = 64 with 8-bit weights (0.737) fits a 1x2 tile.
How small can a keyword-spotting accelerator get, and what does each bit of precision cost?Keyword-spotting accelerator: quantised MLP on Speech Commands, serial-MAC RTL bit-exact against an integer reference, accuracy vs cycles, cells and modelled energy over width and weight bits (stylised).

All 40 Task-specific AI accelerators projects →

Other research topics

Proper scoring & elicitation28 Effort & peer prediction9 Verification games & slashing19 Detection & tolerances14 Markets as routers13 Prediction-market design9 Paying contributors8 Decentralised training systems23 Digital twins & sim-to-real164 LLM chip design loops40 Robot manufacturing loops53 Ambient AI assistant46 Self-replicating manufacturing34 Simulation-first design64 Engineered clothing45 Task-specific AI accelerators40 Algorithmic chip-design loops42 Open-source purpose-built robots44 The software-engineering loop43 Adversarial human imitation41 Open-source residential construction41 Autonomous multi-angle capture34 Sensing, imaging and 3D31 Energy supply and dense energy storage39 Robot societies among people31 Agentic co-design of low-cost robots27 Laundry-hanging robots25 Machine learning on qualitative data23 Robots on the building site27 Civilizations & group psychology1 History & human sciences1

Produced with AI assistance (Claude). Results are from the code in this repository and are simulated or stylised unless the abstract says otherwise; nothing here implies affiliation with or endorsement by any lab, researcher or company cited. MIT licensed.