“The EDGE AI FOUNDATION is the perfect community for us to collaborate on what the future of efficient edge AI hardware looks like. Edge AI is the layer everything else is built on: it determines what machines can sense, decide and act on in the physical world. We’re glad to be building alongside the people advancing it. Heronic is architecting an efficient future for AI chips through automated, bespoke accelerator design. We’re looking forward to sharing knowledge and building partnerships to realise that future.” – Alexander Montgomerie-Corcoran, CEO and Founder
Founded by researchers from Imperial College London, Heronic is developing technology to automate the design of bespoke AI semiconductor hardware. The team has been researching hardware acceleration for AI since 2019, with work spanning object detection, human action recognition, image classification and other compute-intensive applications.
At the center of Heronic’s approach is a simple idea: the future of AI compute may not be one-size-fits-all.
From General-Purpose Compute to Application-Specific AI Hardware
Modern AI accelerators are extraordinarily powerful, but raw peak performance does not necessarily translate into efficient utilization for every workload.
Heronic approaches this problem at the architecture level.
Its MOSAIC — Multi-Objective Synthesis & AI Compiler — platform takes an AI model together with hardware constraints such as performance targets, power budgets, physical area, process technology and available IP. It then explores design permutations to generate a hardware architecture optimized for those specific requirements.
The goal is to automate a process that traditionally requires substantial semiconductor engineering effort: moving from an AI application and its requirements toward specialized, deployment-ready hardware.
For Edge AI, where power, latency, cost and physical footprint are often just as important as raw TOPS, that shift toward application-specific architectures can be particularly important.
AI Designing the Hardware That Runs AI
What makes Heronic especially interesting is the role of automation throughout the semiconductor design process.
MOSAIC combines AI models and physical constraints with techniques including generative design, reinforcement learning and predictive scaling to explore hardware architectures at machine speed.
Customers can specify their trained model, FPGA or ASIC technology target, and Power, Performance and Area objectives. MOSAIC can then estimate resource utilization, power, throughput and latency before producing customized RTL and netlists for downstream implementation.
In other words, Heronic is applying AI not only on semiconductor hardware, but also to the process of designing that hardware.
Putting Specialized Edge AI Into Practice
Heronic is already applying this approach to real-world Edge AI workloads.
Its IRIS video enhancement suite targets applications including super resolution, denoising, frame-rate conversion and HDR tone mapping. Heronic’s IRIS-SR super-resolution technology, for example, performs real-time AI video upscaling on FPGA hardware without requiring a conventional GPU or CPU inference pipeline.
The company is targeting applications across industrial automation, IoT, automotive, telecommunications, broadcast and professional AV — environments where deterministic latency and efficient hardware utilization can matter enormously.
Heronic has also demonstrated its architecture through industry benchmarks and collaborations. Its team previously achieved leading results on the MLPerf Tiny benchmark, and Heronic has worked with Altera on FPGA-based AI acceleration. More recently, the company partnered with ONE WARE to create an automated path from training data and application-specific AI models to deployment-ready FPGA RTL.
We’re excited to welcome Heronic Technologies to the Edge AI Foundation and look forward to connecting their work in automated semiconductor design, FPGA acceleration and application-specific compute with the broader Edge AI community.
Welcome to the EDGE AI Foundation, Heronic Technologies.