Memory Constraints Drive dorsaVi’s Push Toward Advanced 22-nm Neuromorphic Platform

Context: AI Memory Bottlenecks and Industry Trends

As artificial intelligence workloads expand rapidly, traditional memory architectures are increasingly strained by what industry insiders call the “memory wall” and “power wall.” These bottlenecks arise because AI systems demand vast memory bandwidth and capacity, leading to high energy consumption and latency due to frequent data shuttling between processors and external memory. This challenge is particularly acute in edge and ultra-edge devices such as robotics, drones, and autonomous systems, where power, size, and thermal constraints are critical.

Against this backdrop, dorsaVi has embarked on a development program to create a resistive RAM (RRAM) platform at the advanced 22-nanometre (nm) technology node. This initiative aims to deliver a memory solution that is faster, denser, and more power-efficient than legacy nodes, enabling AI computation closer to the data source and reducing reliance on external memory bandwidth.

Progress in RRAM Device Evaluation

dorsaVi recently received its initial RRAM test wafers fabricated at the 180-nm node and has commenced early-stage device characterisation. This phase focuses on assessing device performance, material interfaces, and integration feasibility under manufacturing conditions. Insights gained will inform the company’s roadmap toward a commercially viable 22-nm RRAM platform, which promises significant improvements over the current 40-nm benchmark devices developed in collaboration with NTU.

The 22-nm RRAM targets include reduced write voltage (below 2.0 volts), lower latency (down to 100–200 nanoseconds), and enhanced endurance and retention, all critical for battery-powered and always-on AI systems. Notably, the platform is designed to support compute-in-memory operations, a key feature for neuromorphic computing architectures that mimic brain-like processing to achieve ultra-low power consumption and real-time responsiveness.

Strategic Alignment with Neuromorphic and Edge AI

dorsaVi’s RRAM development is tightly integrated with its neuromorphic computing portfolio, which leverages emerging memory technologies to perform analog-style computation and local learning directly within memory arrays. This approach reduces data movement, a major energy drain in conventional architectures, and enables AI inference and control tasks to be executed on-device without constant cloud connectivity.

The company envisions its 22-nm RRAM platform powering next-generation ultra-edge devices that must process multiple sensor inputs in real time under stringent power and thermal limits. This includes robotics, drones, and autonomous platforms where compact, high-speed, and low-latency memory architectures are essential.

Market Implications and Future Outlook

With global memory supply chains under pressure from surging AI infrastructure demand, dorsaVi’s move toward advanced-node RRAM positions it to address critical industry constraints. The shift toward in-memory and neuromorphic computing architectures is gaining momentum as companies seek to improve performance per watt and reduce dependency on scarce, expensive memory bandwidth.

CEO Mathew Regan emphasised that while much AI investment focuses on large data centres, the future growth phase will increasingly embed intelligence at the edge. dorsaVi’s technology roadmap aims to capitalise on this trend by delivering efficient, manufacturable memory solutions tailored for distributed AI applications.

The company will continue to advance its RRAM and neuromorphic programs through further development and commercialisation milestones, keeping the market informed as progress unfolds.