On-Device AI Features, Shipped and Verified on Your Hardware
Run your detection AI on a single MCU: optimized BOM and memory use, detection that keeps working when the unit is moved, and model care after launch. Deepest expertise on Renesas RA, RZ and RX; other Arm-based platforms supported.
Derived from Greek γνώμων (gnōmōn), literally meaning "one who knows or examines".
Like the gnomon—the pillar of a sundial that casts the shadow to reveal time and direction—we provide the architectural vision and technical clarity that guide hardware forward.
革新的な技術で「人を中心とした」スマートな社会を創造する
To make smart Human-Centric society enabled by innovative technologies.
Technology serves people. We put detection AI on small chips that work quietly in care rooms and buildings, protect privacy by keeping images on the device, and are maintained long after launch.
What We Deliver
Hardware Proof
Your detection model running on the chip you chose, within the latency and memory targets you set. On-device vision today; camera + radar fusion next.
Verification
Every feature verified on your own board, with hardware-in-the-loop regression and scenario tests that run again on each change.
Lifecycle
Sustaining engineering after launch, plus model care that keeps detection models up to date as field conditions change.
Solutions
On-Device Detection for Camera and Sensor Products
For device makers with a camera or sensor product who need it optimized for cost, memory and reliability, with specialist support working alongside their own team.
Care Monitoring
Presence, fall and bed-exit detection for care facilities and hospitals, with images kept on the device.
Facility Security
Intrusion and loitering detection that keeps working in the dark, in glare or when the unit is knocked out of position.
Where we help
- BOM and memory use to optimize for the target price
- Detection drifts when the unit is moved or the room changes
- On-device processing that keeps images and identities private
- Extra specialist hands for porting, cost optimization or verification
Building a different camera or sensor product? The same approach applies.
Engineering in Practice
Featured R&D
Working builds on a single Renesas RA8P1, measured on the board.
Real-Time Pose Estimation
Live body tracking on one microcontroller: camera to 17-point skeleton entirely on the device, with no PC, cloud or network. Moving inference to the Ethos-U55 NPU cut the AI step from 155 ms to 12 ms, so the skeleton updates on every camera frame with about 70% of frame time to spare. Images never leave the device; only body points do.
Conversational Hand Gestures
Two AI workloads on one chip: hand-gesture recognition on the Ethos-U55 NPU and an 8-bit language model on the Cortex-M85, scheduled with fixed µT-Kernel priorities so the camera never drops a frame. A child converses through hand shapes alone, and no camera data leaves the device.
Selected client work (under NDA)
Client and product names are withheld to protect confidentiality — all engagements are covered under NDA.
Embedded vision memory optimization — zero-copy DMA capture and TFLM workspace budgeting under a 2 MB SRAM ceiling on RA8P1.
RA8 architecture R&D — multi-phase evaluation of peripherals, DSP and Helium/NPU acceleration, delivered with verification test suites.
Capabilities
On-Device AI & Vision
- Renesas RA8P1 (Ethos-U55 / Helium)
- int8 Quantization
- Memory-Lean Model Deployment
- TensorFlow Lite Micro / RUHMI
- CMSIS-NN / CMSIS-DSP
Sensor Fusion
- Camera: MIPI CSI-2 / Parallel
- 60 GHz Radar
- Thermal / ToF / IMU
- Decision-Level Fusion & Gating
- SPI / I2C / UART
Verification & Test
- HIL Regression Rig
- Scenario-Based Testing
- On-Device AI Regression
- JTAG / SWD Debugging
- Logic Analyzer
- Memory Footprint Profiling
Lifecycle & Model Care
- OTA Model & Firmware Updates
- Field Performance Monitoring
- Model Re-tuning
- Sustaining Engineering
Legacy MCU Migration
- Renesas RA / RZ / RX Porting
- Bare-Metal to RTOS
- BOM Cost Optimization
- Renesas FSP / e² studio
Platforms & Foundations
- Renesas RA / RZ / RX
- Other Arm Cortex-M / Cortex-A MCUs & SoCs
- FreeRTOS / µT-Kernel / Zephyr
- Embedded Linux (RZ)
- C / C++ / Python
Our Engineering Philosophy
Certainty on Real Hardware
A device feature is only done when it is proven on the chip, verified on your own board, and maintained after launch. We help with each of these steps, with a clear scope and price agreed up front.
How We Work
Three steps, each with a clear scope and price agreed up front.
Feasibility Study
2–6 weeks, with a clear scope and price agreed up front. Can your model run on the chosen chip within the latency and memory targets you set? You get a measured answer and a build plan.
Feature Project
Clear scope and price agreed up front. The feature integrated, optimized and verified on your board, with a hardware-in-the-loop regression rig.
Support Retainer
Annual. Sustaining engineering and model care that keep the feature running well in the field.
Running an ageing MCU? Ask about a legacy MCU migration assessment, with a clear scope and price agreed up front.
Already committed to another platform? We work on the silicon you've chosen, scoped after a feasibility study.
Who You'll Work With
Our Team
Jun Sato (佐藤潤)
Sato leads Gnomons' market strategy, technology consulting, and partnership development from the Tokyo headquarters, connecting the team's engineering capabilities with client opportunities across Japan's embedded, IoT, and edge AI sectors.
Tu Thai (トゥ・タイ)
Tu owns solution architecture and delivery strategy at Gnomons, translating client requirements into buildable embedded systems across the Tokyo and Ho Chi Minh City offices, working closely with clients from initial scoping through delivery and beyond.
Duy Phan (ズイ・ファン)
Duy leads firmware design and hands-on implementation across Renesas silicon platforms, with deep expertise in RTOS design, memory-constrained firmware, and TinyML deployment, delivering production-grade code from prototype through deployment.
Global Presence
Our Offices
Gnomons Co., Ltd.
3-13-5-302 Osaki, Shinagawa-ku
Tokyo 141-0032, Japan
Tel: (+81) 90 9959 0258
1-20-8-604 Osaki, Shinagawa-ku
Tokyo 141-0032, Japan
Tel: (+81) 3 6417 9103 / (+81) 90 9959 0258
Email: services@gnomons.com
Gnomons Vietnam Co., Ltd.
Suite 320, 3rd Floor, Indochina Park Tower
4 Nguyen Dinh Chieu, Tan Dinh Ward
Ho Chi Minh City, Vietnam
Tel: (+84) 28 6288 2040
Email: info@gnomons.com
Renesas distributors and channel partners — we welcome partnership inquiries; reach out anytime.
Start A Conversation
Ready to Build Your Next Edge System?
We are currently accepting new consulting engagements and software development projects. Share your hardware specs, timeline, and core requirements with us.