Literal Labs published a piece by Alex Yakovlev for Designing Electronics North America about MERIT, a novel metric capturing predictive quality and inference energy cost.

Powerful AI. Tiny hardware. Zero GPUs. Literal Labs has pioneered logic-based AI and invented Logic-Based Networks (LBNs). They're a fundamentally different approach to AI that runs on the CPUs, MCUs, and 32-bit semiconductors your products already use. They have no requirement for GPUs, no requirement for cloud connectivity, and certainly no requirement to rip out your existing silicon. LBNs are AI models that solve today's AI bottleneck: the hardware. Independent benchmarks tell the rest of the story. In the MLPerf Tiny anomaly detection benchmark, Literal Labs' LBNs ran 54× faster and consumed 52× less energy than neural networks. And they did so on sub-$5 ARM-based hardware. They've also been deployed on battery-powered IoT hardware. That changes everything about what edge AI can do. You can deploy AI into IoT sensors running on a coin-cell battery. Or build predictive maintenance into factory controllers. Or add decision intelligence to embedded systems that have never run a neural network because, until now, they couldn't. Literal Labs' training platform automates the journey from raw data to a deployment-ready LBN, optimising for your specific hardware, your dataset, and your use case. Models ship as a lightweight C SDK for chip embedding, or as an API via Literal Labs' Managed Inference Server. Logic-Based Networks are also natively explainable and fully deterministic, making them the right choice for regulated industries where AI decisions must be auditable and reproducible. Literal Labs was spun out of Newcastle University by world leaders in logic-based AI, Dr. Alex Yakovlev and Dr. Rishad Shafik, and is led by former Arm CPU division VP and semiconductor startup founder, Noel Hurley.