← All research projectsTESSERA
PURDUE UNIVERSITY

TESSERA

Workload-driven heterogeneous NPU exploration

Modern AI workloads need more than one kind of compute tile. TESSERA explores mixtures of large, small, and special-function tiles, connecting workload mapping with modeled latency, energy, and area.

ResearchPurdue UniversityIISWC 2026 · Published
View original figure ↓Publisher page ↗Published PDF ↗
Original figure

The TESSERA exploration framework

TESSERA Figure 4: workload and architecture inputs plus calibration feed a cost-aware compiler, a tile and chip-level simulator, and a design-space-exploration engine.

The published framework takes a workload and architecture configuration, compiles an execution plan, and estimates its cost with a heterogeneity-aware simulator. Calibration supplies the hardware cost models; the DSE engine compares candidate designs and filters the latency, energy, and area tradeoffs. This is the final collaborative framework.

TESSERA · IISWC 2026 · Figure 4 ↗

01 / THE PROBLEM

Which tile mix makes better use of a fixed hardware budget?

A uniform array of MAC-centric NPU tiles works well for some dense neural networks, but newer workloads spend substantial time on vector operations, FFTs, spiking integration, and polynomial evaluation. Adding specialized tiles creates a large design space: tile counts, precision, memory, dataflow, and operator mapping interact. A fair comparison must account for the shared chip-area and memory-bandwidth budget.

02 / THE APPROACH

Compare complete workload–architecture combinations

01

Describe the workload and tile mix

The framework represents AI workloads as operators and explores combinations of Big, Little, and Special-Function tiles. Architectural choices include array size, precision, SRAM, dataflow, sparsity, and support for non-MAC operations such as FFTs and spiking or polynomial primitives.

02

Compile a cost-aware execution plan

The compiler assigns precision, fuses compatible operators, maps the operator graph, and splits work across compatible tiles when the cost model predicts a benefit. Mapping considers completion time and data movement, so peak arithmetic throughput alone does not determine the tile choice.

03

Simulate shared resources and search the design space

Analytical models estimate tile-level execution and chip-level orchestration, including interconnect and shared DRAM bandwidth. Synthesis-based calibration grounds the costs. A stratified sweep followed by genetic-algorithm refinement searches for Pareto-efficient configurations under common area or latency constraints.

03 / MY CONTRIBUTION

The preliminary exploration prototype

I built the preliminary heterogeneous multi-tile NPU exploration prototype as a Purdue course research project, including synthesis-calibrated cost models and controlled architecture comparisons. I am a co-author of the published IISWC 2026 paper.

04 / EVALUATION

Setup and results

The published collaborative framework evaluates 20 workloads, calibrates cost models to a 7 nm node, and cross-validates against NVDLA. Iso-area comparisons hold the chip budget fixed while changing the composition of compute tiles. My contribution was the earlier Purdue course prototype, including synthesis-calibrated cost models and controlled architecture comparisons.

Mean modeled energy savings
46.91%
Best general-purpose heterogeneous design versus the best iso-area homogeneous baseline.
Chip-area budget
≈200 mm²
For the reported Big + Little + Special-Function design.
Evaluation suite
20 workloads
Including dense, vector-heavy, and non-MAC computation patterns.
05 / OUTCOME & SCOPE

A workload-driven choice of tile heterogeneity

The published results show how reallocating a fixed silicon budget can serve a broader workload mix more efficiently than replicating one tile type. These are calibrated simulation and design-space-exploration results. The final framework and its reported gains are collaborative work; they should be distinguished from my preliminary course prototype and from measurements on fabricated silicon.