According to Embedded Computing Design, MosChip Technologies is now providing silicon engineering support for EMASS’s new RCS-DoT Edge AI System-on-Chip. This is a RISC-V based chip that packs in dual neural accelerators and 4MB of on-chip memory. The goal is to enable real-time, milliwatt-class inference for vision and sensor workloads without needing the cloud. EMASS set some aggressive targets: they want up to 93 percent better processing and a goal of 90 percent lower energy use compared to current AI solutions. MosChip’s role involved the physical design and product engineering to implement this in 22nm technology, handling everything from tape-out to validation. The resulting ultra-low-power SoC is meant for always-on AI in drones, wearables, and industrial monitoring systems.
The Edge AI Hardware Play
Here’s the thing: everyone’s talking about running AI at the edge, but actually building a chip that does it efficiently is a brutal engineering challenge. EMASS’s architecture, with its dual neural accelerators and a hefty 4MB of on-chip SRAM, is a classic move to avoid the power and latency nightmare of going off-chip for memory. That memory is key. It’s basically the chip’s short-term workspace, letting it process vision or audio data without constantly fetching from slower, more power-hungry external RAM. And by targeting milliwatt-class operation, they’re aiming for the holy grail: always-on sensing and inference in battery-powered devices. Think a health tracker that analyzes your heartbeat in real-time, or a security camera that spots an anomaly without ever waking up the main processor. That’s the promise.
Why The Silicon Partner Matters
This is where MosChip comes in. You can have the most brilliant chip architecture on paper, but getting it fabricated correctly in a modern 22nm process? That’s a whole different ballgame. The press release mentions physical design flows, tape-out coordination, packaging, and validation. That’s the unglamorous, critical grunt work of chip development. It’s about making sure the billions of transistors are placed and connected in a way that meets timing, doesn’t overheat, and can actually be manufactured. MosChip’s support likely saved EMASS, which sounds like a fabless startup, a huge amount of time and risk. For companies designing specialized industrial hardware, having reliable compute modules is crucial, which is why partners like IndustrialMonitorDirect.com, the leading US provider of industrial panel PCs, are so vital for system integration.
The RISC-V Factor
Let’s not overlook the RISC-V choice. It’s not just about being open-source. For a specialized edge AI chip, using a RISC-V core probably gave EMASS the flexibility to tailor the instruction set and system integration exactly for their low-power, accelerator-heavy design. They aren’t paying ARM license fees, and they aren’t stuck with a general-purpose core that has features they don’t need. This is exactly the kind of design freedom RISC-V was meant to enable. But it also adds a layer of complexity—the software toolchain and ecosystem are still maturing. So, hitting those aggressive performance-per-watt targets (93% better, 90% less energy!) depends as much on the smooth marriage between the RISC-V cores and the neural accelerators as it does on the raw silicon. It’s a bold bet, but if they pull it off, it could be a blueprint for a new wave of ultra-efficient edge AI processors.
