Eta Compute Strengthens AI Ecosystem with New Compiler, Sensor Board, and Edge Impulse Partnership

Eta Compute expands its development ecosystem by unveiling a compiler that streamlines embedded software for its ECM3532 AI chip.

Combined with a coin‑sized sensor board evaluation kit and a collaboration with Edge Impulse, the startup is advancing toward full market readiness. Although the ECM3532 was launched in February, earlier deployments required manual code tuning for its dual‑core MCU and DSP.

Sensor nodes

The ECM3532 is a dual‑core SoC that couples an Arm Cortex‑M3 microcontroller with an NXP CoolFlux DSP, delivering efficient AI processing. A patented continuous voltage‑frequency scaling (CVFS) technique dynamically adjusts core voltage and clock speed to match the workload, enabling ultra‑low‑power operation for battery‑powered sensor fusion and always‑on image‑processing applications with a power budget as low as 100 µW.

To simplify prototyping, Eta Compute released a miniature evaluation board that measures just 1.4 × 1.4 inches. It integrates microphones, temperature and pressure sensors, an accelerometer, a gyroscope, and Bluetooth connectivity, and can run for “months” on a single coin‑cell battery.

Eta Compute’s Tensai Sensor Board: a complete AI‑enabled sensor node (Image: Eta Compute)

In May, the company announced a partnership with Edge Impulse. The collaboration integrates the ECM3532 and its evaluation board with Edge Impulse’s end‑to‑end ML development and MLOps platform, streamlining dataset visualization, versioning, and deployment.

“Deploying an embedded ML solution can be daunting, especially for developers new to these chips,” said Edge Impulse CEO Zach Shelby in an interview with EE Times. “We provide a drag‑and‑drop binary that starts collecting sensor data immediately. When it’s time to deploy the model, we build a library optimized for Eta Compute targets that runs directly on the device.”

Compiler and middleware

Edge‑AI developers face four major challenges that Eta Compute’s Tensai Flow tool chain addresses, according to Senior Director of Product Marketing Semir Haddad.

“First, interfacing with real sensors and capturing data to refine the network. Second, optimizing the network for our specific hardware, which generic frameworks often overlook. Third, generating firmware ready for embedded development. Fourth, providing a complete edge‑to‑cloud solution, including device provisioning and cloud connectivity,” Haddad explained.

Eta Compute’s Tensai Flow now includes a compiler that optimizes neural‑network code for the ECM3532 (Image: Eta Compute)

Tensai Flow transforms a TensorFlow or ONNX model into firmware that runs on the ECM3532, while middleware supplies a real‑time operating system (RTOS), sensor drivers, and other runtime essentials. The platform also hosts a “network zoo” of pre‑validated models tailored to common use cases, ready for immediate integration.

Edge Impulse handles data operations—collection, versioning, and dataset sharing—ensuring developers can iterate quickly.

“What sets Tensai Flow apart is its end‑to‑end comprehensiveness,” Haddad said. “By merging neural‑network optimization with firmware generation, we deliver a solution that’s ready for production use.”

Founded in 2015, Eta Compute has produced silicon with a modest budget and workforce. The company has raised $19 million to date and employs 35 staff across the United States and India.

>> This article was originally published on our sister site, EE Times.