Signaloid, a UK-based computing company, has joined CERN openlab's Heterogeneous Architectures Testbed for advanced computing research and standardization of heterogeneous processor integration.

British computing company Signaloid has joined CERN openlab, the public-private partnership through which CERN evaluates emerging information technologies for scientific computing. The partnership will evaluate Signaloid's distribution-extended compute hardware (UxHw) technology within the CERN openlab Heterogeneous Architectures Testbed, which explores new processor technologies for future scientific computing infrastructures.

This collaboration reflects a broader shift across research organizations worldwide, which are increasingly adopting heterogeneous computing architectures combining CPUs, GPUs, and specialized accelerators to handle demanding workloads more efficiently. As the High-Luminosity Large Hadron Collider (HiLumi LHC) approaches, CERN is evaluating emerging computing technologies to address its rapidly growing computational requirements.

"Heterogeneous architectures are becoming essential for the HiLumi LHC," says Maria Girone, CTO of CERN openlab. "CERN openlab is pioneering a model for evaluating next-generation computing technologies such as Signaloid's distribution-extended compute hardware. The upcoming deployment of Signaloid's hardware and software stack illustrates the kind of architectural innovation openlab was created to evaluate."

The Large Hadron Collider is the world's largest particle accelerator. By colliding protons at nearly the speed of light, CERN's scientists investigate the fundamental building blocks of matter and pursue answers to some of physics's biggest questions. Much of this research depends on Monte Carlo simulation, which allows physicists to compare experimental measurements with millions of simulated particle collisions. Generating these simulations requires repeatedly calculating the same physical processes using different random inputs, making Monte Carlo event generation one of the most computationally-demanding workloads in particle physics.

When the HiLumi LHC begins operation later this decade, the number of recorded collisions will increase dramatically. While this promises unprecedented scientific opportunities, it will place enormous pressure on CERN's computing infrastructure, with projected demand expected to outpace available resources.

Rather than replacing conventional processors, Signaloid's UxHw technology extends heterogeneous computing systems with native support for computation directly on digital representations of continuous probability distributions. With this capability, software running on UxHw can perform calculations directly on probability distributions in a single execution pass, while requiring minimal changes to existing software—eliminating the need for repeatedly executing the same kernel millions of times with different random inputs.

In benchmarking against today's high-performance server platforms, UxHw has demonstrated speedups of up to 2,000× for representative workloads ranging from high-energy physics to regulatory risk modeling for banks, chip design simulations, and robotics. Additional efficiency gains are expected from Signaloid's recently custom ASIC implementations, the first of which taped out in May 2026 using a low-power TSMC fabrication process.

As part of the Heterogeneous Architectures Testbed, CERN and Signaloid will evaluate a representative Monte Carlo event generation workflow based on the Pepper framework for proton-proton collisions producing multiple gluons. The project will assess computational performance, numerical accuracy, and integration effort, helping determine where distribution-extended computing technologies like UxHw can complement existing CPU- and GPU-based scientific computing infrastructure. The joint project aims to identify where distribution-extended computing adds value within the Monte Carlo event generation pipeline and what practical considerations are involved in integrating the technology into existing high-energy physics software.

"The future of high-performance computing will not be defined by a single processor architecture, but by heterogeneous systems that combine specialised hardware for different classes of computation," said Prof. Phillip Stanley-Marbell, Founder and CEO of Signaloid. "We're excited that CERN openlab is evaluating UxHw alongside other emerging computing technologies. Experimental particle physics represents one of the most demanding and exciting environments in which to demonstrate its potential."

Dr. Stefan Roiser, Senior Computing Engineer at CERN, added: "The largest share of LHC computing resources is spent simulating particle collisions. We will explore Signaloid's technology in Monte Carlo event generation, the first step in the simulation chain expected to see substantial cost increases during CERN's upcoming High Luminosity data-taking period. Because event generation relies heavily on multi-dimensional distributions, Signaloid's UxHw technology has strong potential to accelerate this software and help meet the forecasted computing budgets during HiLumi LHC, starting in 2030."

This collaboration reflects a broader trend across high-performance computing. Around the world, governments and research organizations are investing in heterogeneous computing systems capable of combining multiple processor architectures for increasingly complex scientific and AI workloads, including the UK's planned £750 million AI Research Resource (AIRR) heterogeneous supercomputer.

Signaloid develops a computing platform that reduces the runtime and compute infrastructure requirements of workloads ranging from engineering simulations to robotics and physical AI. Founded by Prof. Phillip Stanley-Marbell, former Professor of Physical Computation at the University of Cambridge, the company's compute platforms enable software to operate directly on probability distributions through its distribution-extended compute hardware architecture. Signaloid's technology is available through cloud, on-premises, and edge computing platforms and is used by more than 3,000 developers worldwide.