Nellow provides the SPARK™ Edge AI Processor, a ferroelectric non‑volatile logic‑memory chip that combines compute and storage to eliminate static power draw and data movement. The platform delivers sub‑watt AI inference for transformers, vision and small language models, with zero standby power and an accompanying ORBIT™ SDK for model deployment, OTA updates, and security features.
Funding
Funding not disclosed
Founders
Product
Problem
The energy consumption of CMOS-based microelectronic chips is a growing environmental and industrial concern, limiting performance due to heat and increasing operating costs for data‑center, edge‑AI, and embedded computing applications.
Solution
Nellow’s technology replaces conventional CMOS transistors with quantum‑material, ferroelectric‑based devices that are non‑volatile and combine logic and memory in a single element. This architecture eliminates static power draw and removes the energy‑intensive data movement between separate logic and memory blocks. By operating at ultra‑low switching energies and reduced voltages, the chips achieve orders‑of‑magnitude lower energy per operation, enabling high‑performance AI inference and general computing within sub‑watt power budgets. The company offers the SPARK™ Edge AI Processor together with the ORBIT™ SDK, providing a hardware‑software platform for on‑device transformer, vision, and small‑language‑model inference with zero standby power. Customers can integrate the processor via standard interfaces (PCIe, MIPI CSI‑2, etc.) and manage models through OTA updates, secure enclaves, and real‑time power profiling.
Target Audience
Primary customers are manufacturers of data‑center accelerators, edge‑AI devices, and embedded systems that require high‑performance inference under strict power constraints, such as industrial vision, autonomous robotics, smart infrastructure, and battery‑operated medical equipment.
Features
- Ferroelectric, non‑volatile logic‑memory devices that retain state without power, removing static leakage
- Integrated logic and memory in a single element, eliminating energy‑costly data transfer
- Ultra‑low switching energy and low operating voltage delivering <1 W power consumption for AI inference and zero standby draw
- SPARK™ Edge AI Processor supporting transformers, CNNs, VLMs, and small language models with PCIe, MIPI CSI‑2, ADC/DAC, and serial interfaces
- ORBIT™ SDK with Python‑first APIs, automatic model quantisation, on‑device runtime, OTA updates, and power profiling tools
- Built‑in security features including root‑of‑trust, cryptography, and lifecycle management
- Compatibility with Yocto‑based BSP, FreeRTOS, and bare‑metal environments for flexible deployment