01 – HARDWARE
Enables on-device transformer inference, SLMs, VLMs and vision models, at power budgets impossible with conventional CMOS. Non-volatile, zero standby consumption, with logic and memory co-located in a single device.
Architecture
Cascaded architecture aligned with models execution to enable real-time reasoning applications
Auto-adaptive coarse & fine grain power gating
AI Workloads
Transformers, SLM, VLM, CNNs
Power Consumption
Ultra-low power <1W for reasoning / Zero standby
Technology
Beyond-CMOS NMC & IMC processing
Interfaces
PCIe, MIPI CSI2, ADC/DAC, Serials, xSPI
Security
Root-of-Trust, cryptography, life cycle management
Platform
Yocto-based BSP + RT AI, FreeRTOS
02 – SOFTWARE
A complete developer platform to deploy AI models on the Nellow chip. From model import to hardware-optimised inference with Python-first APIs and CLI tools.
Model Import
Load pre-trained transformers from HuggingFace or ONNX. Automatic quantisation and hardware-specific optimisation.
HuggingFace · ONNX
Vision Pipeline
Pre-built pipelines for object detection, scene classification, OCR and VLM visual reasoning. Camera-in, inference-out API.
Object Detection · VLM · OCR
Runtime & Scheduler
Optimised inference runtime with model multiplexing, dynamic batching and power-mode scheduling.
Multi-model · Low-latency
SLM Reasoning
Run small language models on-device for context-aware reasoning, anomaly detection and decision logic, fully offline.
SLM · On-device · Offline
Power Profiler
Real-time energy measurement with per-layer power breakdown. Idle vs. active characterisation and optimisation suggestions.
Energy · Profiling · Debug
Edge Deployment
OTA model updates, fleet management, and secure enclave support. Deploy to embedded Linux, RTOS or bare-metal.
OTA · RTOS · Linux
03 – APPLICATIONS
From industrial vision to medical devices, Nellow’s edge AI platform enables applications with reasoning previously constrained by power budgets.
Industrial Edge AI
Defect detection, quality control and process monitoring with real-time VLM inference on production lines.
Autonomous Robotics
Scene understanding and navigation reasoning for drones, mobile robots and collaborative arms.
Smart Infrastructure
Traffic analysis, crowd monitoring, environmental sensing, always-on at near-zero power draw.
Medical Devices
Portable diagnostics with AI-assisted image analysis, battery-constrained, privacy-preserving, offline.
04 – SDK PREVIEW
From model to inference in minutes, not weeks. No hardware expertise required with ORBITTM SDK.
import nellow_sdk as nw
# Load a vision-language model on the Nellow chip
model = nw.load_model(« vlm/nano-2b », device=« nellow:0 »)
# Capture a frame and run inference
frame = nw.capture(source=« cam0 »)
result = model.infer(frame, prompt=« Describe any anomaly »)
print(result.text, f »— {result.power_uw:.1f} µW »)
05 – ROADMAP
From industrial vision to medical devices, Nellow’s edge AI platform enables applications with reasoning previously constrained by power budgets.
2029
Low Power Inference
Decision-making Edge AI chips
2032
Low Power Training
Low voltage learning for 10x energy efficiency gain
2035
Low Power Logic
100x energy efficiency gain
5x reduction of the wafer area