Signaloid's UxHw distribution-extended compute hardware — which the Cambridge-based company says has demonstrated speed-ups of up to 2,000× against today's high-performance server platforms — will be evaluated inside CERN openlab's Heterogeneous Architectures Testbed, under a collaboration announced Oct. 1, 2026. The joint project will run a representative Monte Carlo event generation workload built on the Pepper framework and measure computational performance, numerical accuracy, and integration effort.

The timing is not incidental. CERN's computing budget is heading for a collision of its own. When the High-Luminosity Large Hadron Collider (HiLumi LHC) begins data-taking, projected computing demand is expected to outpace available resources, and the first step of the simulation chain — event generation — is where costs are forecast to climb steeply.

What does UxHw actually change?

Signaloid's approach does not replace conventional processors. Instead, UxHw extends heterogeneous systems with native computation directly on digital representations of continuous probability distributions. In classical Monte Carlo workflows, the same kernel executes millions of times with different random inputs — the sampling approach that dominates particle-physics simulation today.

Software running on UxHw performs calculations directly on probability distributions in a single execution pass, requiring minimal changes to existing software, according to the company. The claimed 2,000× speed-ups span representative workloads from high-energy physics to regulatory risk modeling for banks, simulations used in chip design, and robotics.

The first of Signaloid's custom ASIC implementations taped out in May 2026 on a low-power TSMC fabrication process, with additional efficiency gains expected from these silicon implementations.

Why is CERN evaluating it now?

Monte Carlo simulation underpins much of the LHC's science. Physicists compare experimental measurements against millions of simulated particle collisions, generated by repeatedly calculating the same physical processes with different random inputs. This makes Monte Carlo event generation one of the most computationally demanding workloads in particle physics.

"The largest share of LHC computing resources is spent simulating particle collisions," said Dr. Stefan Roiser, Senior Computing Engineer at CERN. "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."

Roiser points to the structural fit: because event generation relies heavily on multi-dimensional distributions, UxHw has strong potential to accelerate this software and help meet forecasted computing budgets during HiLumi LHC, starting in 2030.

CERN openlab, the public-private partnership through which CERN evaluates emerging information technologies for scientific computing, is positioning itself as the proving ground. "Heterogeneous architectures are becoming essential for the HiLumi LHC and CERN openlab is pioneering a model for evaluating next-generation computing technologies such as Signaloid's distribution-extended compute hardware technology," said Maria Girone, CTO of CERN openlab. "The upcoming deployment of Signaloid's hardware and software stack at CERN openlab illustrates the kind of architectural innovation openlab was created to evaluate."

How does the evaluation work?

The testbed project will assess three dimensions:

• Computational performance on the Pepper-based workflow for proton-proton collisions producing multiple gluons;
• Numerical accuracy of distribution-extended computation versus conventional sampling;
• Integration effort — the practical considerations of fitting UxHw into existing high-energy physics software stacks built for CPUs and GPUs.

The stated goal is to identify precisely where distribution-extended computing adds value within the Monte Carlo event generation pipeline, and where existing infrastructure remains sufficient.

A heterogeneous computing trend, not a one-off

The collaboration sits inside a broader shift. Research organizations worldwide are adopting architectures that combine CPUs, GPUs, and specialized accelerators for demanding workloads, and governments are funding heterogeneous systems for scientific and AI computing — including the UK's planned £750 million AI Research Resource (AIRR) heterogeneous supercomputer.

"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, Signaloid's founder and CEO. "Experimental particle physics represents one of the most demanding and exciting environments in which to demonstrate its potential."

Stanley-Marbell, formerly Professor of Physical Computation at the University of Cambridge, founded Signaloid around the UxHw architecture. The company's platform is available through cloud, on-premises, and edge deployments and claims more than 3,000 developers worldwide.

The commercial stakes extend well beyond Geneva. The same probability-distribution workloads that dominate LHC simulation appear in bank risk modeling, chip-design simulation, and robotics — and Signaloid's 2,000× benchmark claims will now face scrutiny in one of the world's most demanding computing environments. Whether UxHw earns a place alongside CPUs and GPUs in CERN's HiLumi-era infrastructure will hinge on the testbed's accuracy and integration findings over the coming months.