MicroQuant v1
Customer-specific GA · available on demand

Embedded inference built for your product.

MicroQuant v1 turns bounded, statically shaped integer graphs into deterministic ABI-v6 C. The promoted schemes are W8, PCQ4, SPQ4, KARQ, and mixed precision, with explicit limits, caller-owned workspace, and no runtime model parser.

Discuss your implementation Read the support contract
Current envelope ABI 6 5 schemes Portable C
MicroQuant compilerMQ-V1
ABI 6/6 compiler / runtime
Schema closedfail closed
W8–4bpromoted schemes
No heapruntime execution
256maximum graph operators
1–4graph inputs
1–4graph outputs
256 KiBmaximum caller workspace
Measured ABI-v6 reference result

Smaller firmware.
Faster inference.

Speech Commands v2 DS-CNN measured on an ESP32-S3 at 240 MHz. MicroQuant PIE is compared with portable TFLite Micro, ESP-NN, and one generic-SWAR W8 control.

1.28×KARQ P1 throughput vs ESP-NN
−91.0%KARQ P1 product flash vs ESP-NN
−45.5%MicroQuant peak SRAM vs ESP-NN

Speech recognition benchmark

Full 4,769-record top-1 accuracy; exact-board p50/p99 latency and linked memory accounting.

ESP32-S3 · ESP-IDF 6.0.2
Configuration Backend Top-1 p50 latency p99 latency Product flash Firmware flash Peak SRAM
TFLite MicroPortable91.99%430.017 ms430.064 ms262,944 B262,944 B55,936 B
TFLite Micro + ESP-NNESP-NN91.99%18.281 ms18.361 ms291,424 B291,424 B56,400 B
MicroQuant W8generic-SWAR92.24%232.601 ms232.605 ms34,678 B201,216 B30,712 B
MicroQuant W8PIE92.24%15.441 ms15.446 ms34,909 B201,424 B30,744 B
MicroQuant PCQ4PIE91.36%15.734 ms15.737 ms24,797 B191,312 B30,744 B
MicroQuant SPQ4 G8PIE91.70%37.646 ms37.649 ms34,349 B200,864 B30,744 B
MicroQuant KARQ P1PIE91.40%14.288 ms14.291 ms26,229 B192,752 B30,744 B
MicroQuant KARQ P4PIE90.14%17.007 ms17.010 ms35,049 B201,568 B30,744 B
MicroQuant MixedPIE91.03%30.407 ms30.414 ms31,069 B197,584 B30,744 B

Accuracy uses the frozen feature task set; latency uses one reset-isolated 100-endpoint physical run per cell. Every MicroQuant output matched its host result bit for bit and every measured cell reported zero inference heap growth. This reference campaign excludes microphone capture, feature extraction, and energy.

The immutable ABI-v6 campaign is bound to source commit c970e013. Customer-specific GA remeasures the delivered model, generated artifacts, and firmware, so results are never transferred blindly to a different product.

Download the machine-readable benchmark
Promoted v1 coverage

A narrow contract.
A complete failure boundary.

Every accepted model is validated before output is committed. Unsupported shapes, operators, quantization, targets, and malformed inputs fail with structured diagnostics.

Dense Conv2D Depthwise Conv2D Add Average Pool Global Average Pool Max Pool Identity Flatten ReLU Clamp
Bounded ingestion

Four explicit source formats.

Canonical ABI-v6 JSON, source-authorized JSON, quantized ONNX QLinearMatMul, and signed-int8 TFLite fully connected graphs enter the same typed compiler path.

Download machine-readable status
Customer-specific GA delivery

A production path
qualified for your exact product.

MicroQuant v1 is available through customer-specific GA engagements. We bind the model, generated artifacts, firmware, target toolchain, verification, and support to the exact implementation instead of offering a one-size-fits-all generic binary.

Delivery modelGAcustomer-specific
01

Compiler + runtime

ABI-v6 production foundation

Deterministic generated C, strict C/C++ integration, lifecycle checks, and numerical proof replay form the common v1 foundation.

02

Target integration

ESP32-S3 ready

The exact ESP-IDF toolchain and generated-source path are ready for customer qualification; additional microcontrollers are evaluated on demand.

03

Implementation validation

Bound to your product

Board measurements, received-byte parity, resource budgets, and integration tests are produced for the customer's exact model and firmware.

04

GA delivery

Available on demand

Each engagement completes the production signing, commercial terms, acceptance, support, and lifecycle package for that customer-specific release.

Read the machine-readable status Read the evaluation terms
Integration path

Model to generated C.
No hidden runtime.

The development workflow separates admission, compilation, received-byte verification, and firmware ownership so each boundary can be tested independently.

  1. 01

    Preflight

    Validate the model format, graph envelope, quantization, limits, target, and proof prerequisites without writing output.

  2. 02

    Compile atomically

    Create deterministic generated C, manifests, integrity identities, and an explicit manual source list.

  3. 03

    Verify and integrate

    Verify the received kit, compile the declared files, provide caller-owned buffers, and retain target acceptance as a separate activity.

Technical evaluation

Does your graph fit the v1 envelope?

Share the model format, graph shape, target, and constraints. We will qualify the fit and scope a customer-specific GA implementation for your product.

First stepContract-bound model review
✓

Request received.

We will follow up at .