Highlights

  • Nanoveu's (ASX:NVU) Wholly Owned Subsidiary EMASS has published new benchmark results for its ECS-DoT-22 chip, using purpose-built AI models optimised for the chip's deep-learning accelerator.
  • ECS-DoT-22 is claimed to demonstrate lower energy per inference than all published competitor results on the MLPerf Tiny benchmark across all four standard workloads: keyword spotting, visual wake words, image classification, and anomaly detection.
  • Sub-microjoule inference has been achieved for keyword spotting and anomaly detection, with higher accuracy than the reference models on every workload tested.
  • ECS-DoT-22 evaluation kits are available through distributor DigiKey, with the EMASS integrated development environment and AI compiler due for release in November 2026.

Nanoveu Limited (ASX:NVU), a technology company with operations spanning advanced semiconductor, drone, and materials sciences, has announced new benchmark results for the ECS-DoT-22 chip developed by its wholly owned subsidiary, Embedded A.I. Systems Pte Ltd (EMASS). The results use models purpose-built for the chip's deep-learning accelerator (DLA), and EMASS claims they demonstrate lower energy per inference than all currently published competitor results across all four standard MLPerf Tiny workloads.

Latest Development

The benchmarks cover four workloads: keyword spotting, visual wake words (VWW), image classification, and anomaly detection. These are the standard workloads specified by MLCommons for the MLPerf Tiny benchmark, an industry reference for assessing AI inference performance on microcontroller-class devices. Each new EMASS model was required to meet or exceed the accuracy of the standard reference models on the respective official test datasets before its energy and latency were measured.

EMASS reports that at 50 MHz operating frequency, the new accelerator-native models achieve sub-microjoule energy per inference for keyword spotting and anomaly detection. For image classification and visual wake words, the energy levels are also below every currently published result from active vendors with MLCommons submissions, based on EMASS's comparisons with published figures from Syntiant, STMicroelectronics, Silicon Labs, Renesas, Asygn, Ambiq, and BrainChip.

Technical Significance

The distinction between energy per inference and system-level energy is important: EMASS reports energy drawn by the chip's core Supply rail during one inference, excluding other board-level components. The company notes that at ten keyword-spotting inferences per second, the AI compute cost amounts to approximately 6 microwatts — a level low enough for the AI processing to become essentially a minor contributor to overall device power budgets in battery-operated sensors and wearables. This is compared to what the company describes as a substantially higher figure for the nearest published competitor at the same workload.

EMASS also reports that its new models outperform MLPerf Tiny reference models in accuracy on every workload, clearing the official quality thresholds by material margins, while achieving latency improvements of up to nine times on the same chip at the same clock speed.

Product Availability and Commercialisation

ECS-DoT-22 evaluation kits are described as available through DigiKey, following a distribution agreement announced on 10 November 2026. The EMASS integrated development environment (IDE) and AI compiler are expected to be released in November 2026, and the benchmark models, projects, and host scripts behind the announced results are planned to be made available with the IDE release, enabling customers and third parties to reproduce the results independently.

The company positions ECS-DoT-22 for markets where always-on AI inference at ultra-low power is a constraint — including hearables, wearables, smart devices, industrial sensing, and robotics. Mark Goranson, Nanoveu's CEO of Semiconductor Technologies, noted that energy per inference is the first question engineering teams ask at Design-In stage.

Company Background

EMASS is a fabless semiconductor company within Nanoveu's portfolio. Nanoveu also operates Spinoff Robotics, a Singapore-based aerial robotics Business with drone platforms for high-payload operations, and EyeFly3D, a glasses-free 3D display technology. The company also holds interests in Nanoshield, a protective coating technology.

Sector Context

Edge AI semiconductors represent a growing segment within the broader semiconductor industry, as processing moves closer to sensors and away from centralised cloud infrastructure. Power consumption is a primary constraint for battery-operated edge devices, and suppliers capable of demonstrating measurably lower energy per inference have a potential Competitive Advantage in design-win engagements.

Growth Drivers

The release of evaluation kits through DigiKey and the forthcoming IDE launch are intended to lower the barrier for customer evaluation of ECS-DoT-22. If independent verification of the EMASS performance claims proves consistent, the chip may attract interest from OEM design teams seeking differentiated energy performance for next-generation edge AI products.

Risks and Challenges

The benchmark results as published by EMASS have not been independently verified by MLCommons, and the company acknowledges that comparisons involving third-party results use different benchmark rounds and include both verified and unverified submissions. Commercial adoption remains subject to customer qualification, integration processes, and procurement decisions. The fabless semiconductor business model is capital-intensive during development, and translating benchmark performance into commercial design wins typically takes multiple product cycles.

Investor Considerations

For investors in the deep technology and semiconductor sector on the ASX, Nanoveu's EMASS represents an asset that is at an early stage of commercial engagement, having demonstrated benchmark performance on its evaluation platform. The critical next steps involve converting benchmark results into commercial design-wins with original equipment manufacturers — a process that typically takes considerably longer than benchmark publication. The DigiKey distribution agreement and the forthcoming IDE release represent the first infrastructure for independent evaluation, which is a prerequisite for any formal qualification process. Investors should note that the semiconductor industry operates on extended sales cycles for new chip designs, often measured in years rather than months. Additionally, Nanoveu's portfolio spans multiple technology areas including drones and 3D display technologies, meaning the semiconductor segment is one of several activities being developed simultaneously within a small company.

Key Insights

Nanoveu's EMASS subsidiary has published benchmark results claiming the lowest publicly available energy per inference across all standard MLPerf Tiny workloads for the ECS-DoT-22 chip, using accelerator-native models. With evaluation kits now available through DigiKey and the EMASS IDE expected imminently, the company is moving from performance claims toward customer qualification. The near-term focus will be on independent verification and design-in engagements with OEMs in the sensor, wearable, and industrial IoT spaces.