At a glance
| Booth | I3000 |
| Country | GB |
| Website | hartleyultrafast.com |
Company profile
HARTLEY ULTRAFAST LTD is a Bristol, UK deep-tech hardware company developing AI decision-making silicon with light for latency-critical applications. Its optoelectronic architecture combines silicon photonics and electronics and computes small, dense neural networks directly in the analogue photonic domain. The architecture is designed for nanosecond-class inference rather than microsecond-scale inference and provides a standard UCIe chiplet interface for integration into existing CPU, FPGA and SoC designs without redesigning the host system. Hartley's coprocessors solve entire neural networks using a fusion of photonics and electronics and are positioned for real-time systems, enabling infrastructure and adversarial environments where latency is critical. Physical integration is designed to be standard on-board or in-package, while software integration is provided through RTL and Python with standard SWAP requirements. The company has filed four patents on its core architecture and has a founding team that includes five PhDs in photonics, physics and computer science. Hartley targets ultrafast decision-making applications in high-frequency trading, electronic warfare, communications and forward error correction, industrial process control and AI acceleration. Its target market is $162B in 2025, growing to $450B by 2030, for application areas where each nanosecond of latency has measurable value. The company is building manufacturing, packaging, integration and field-trial partnerships to move nanosecond-class AI into production silicon. Hartley also participates in Altera's global ASAP partner network, coupling its optoelectronic neural-processing units with an FPGA platform for defence, industrial control and capital-market applications. Hartley's design philosophy starts from first-principles hardware and fixed-point computation while using optoelectronic, analogue processing to reduce decision latency in systems where delay has operational consequences.
Exhibits
Babbage is HARTLEY ULTRAFAST LTD's ultrafast inference proof of concept and an optomechanical architectural demonstrator for its light-based neural-computing approach. Babbage is presented as a photonic neural-network computer that demonstrates AI inference in the nanosecond domain without relying on microsecond-scale digital inference timing. Clockwork is the company's first production timing chip and is scheduled to tape out on an advanced RF process in Q2 2026. Hartley also shows Clockwork silicon prototypes as part of its hardware development path. The architecture exposes a UCIe chiplet interface for CPU, FPGA and SoC integration and supports standard physical integration on-board or in-package. Software integration is supported in RTL and Python, allowing the optoelectronic processing unit to be incorporated into existing system-development workflows. The exhibit scope therefore centers on ultra-low-latency optoelectronic AI inference, analogue-photonic neural-network computation, UCIe chiplet integration and nanosecond-scale decision processing for time-critical systems. The hardware development set also includes a Hartley prototype backplane, complementing Babbage and Clockwork prototypes as integration hardware for ultrafast systems. Babbage is also used as the architectural proof of concept through which key design features and performance results for the optoelectronic approach are presented.
Capabilities and products
- light-powered AI decision-making silicon
- analog photonic neural-network computation
- nanosecond-class AI inference
- UCIe chiplet integration with CPU/FPGA/SoC
- RTL and Python software integration