Phillip Stanley-Marbell is the founder and CEO of Signaloid. Up until September 2025, he was a full Professor (Chair of Physical Computation) in the Electrical Engineering Division of the Department of Engineering at the University of Cambridge, where he led the Physical Computation Laboratory. He completed his PhD at Carnegie Mellon and has held full-time positions across industry and academia, including at Bell Labs, IBM Research Zurich, Apple, and MIT.

Tell us about your company.

Signaloid provides computing platforms targeting applications that involve, or can benefit from, stochastic methods. Our core technology, which we call “distribution-extended compute hardware” (UxHw), allows algorithms for these target applications to achieve 10-fold to 100-fold performance improvements, often with minimal changes. The applications include, but aren’t limited to,

• Robotics and physical AI
• Industrial automation
• Engineering simulations
• Pharmaceutical R&D
• Trustworthy AI
• Bayesian machine learning

The idea behind Signaloid was born while I was a researcher at MIT CSAIL and was inspired by lessons I learned working at Apple and at IBM before that. After leaving MIT to join the University of Cambridge as a professor in 2017, I built a new team (the Physical Computation Laboratory) to explore the idea further, and then bootstrapped the company during the pandemic.

We’ve now successfully shipped two families of products based on our core UxHw technology.

• A high-performance enterprise platform, for which we are working with companies like Boeing, CERN, and major international banks.
• A hardware module, which packages the same core technology for robotics and industrial automation. We have distribution partnerships with Bosch and Mouser, and ongoing pilots with Infineon, ABB, and the German Aerospace Agency (DLR), among others.

In May 2026, we taped out a new ASIC which will go into our future hardware modules. Along the way, we’ve built a significant amount of intellectual property, with over 100 patent filings across the US, EU, Japan, China, and Taiwan.

2. What problems are you solving?

Numbers are an essential foundation of human civilization. From commerce and robots to pharmaceutical research simulations, computers allow us to harness numbers for our lives and livelihoods. But the way computers handle numbers has largely remained unchanged for decades, leading to increasing pressure to improve compute efficiency beyond what semiconductor process scaling alone can deliver.

 

Signaloid extends how computers work today, to better match the workloads of the future. Our computing platform maintains metadata (discretizations of continuous probability distributions) paired with all floating-point values inside programs as they run, without changing program behavior. This additional data, together with efficient propagation of the metadata through programs as they run, allows software running on our platform to be made faster with minimal additional tweaks. Often, it enables 10- to 100-fold performance improvements.

 

To give a few concrete examples, our enterprise platform provides 400x speedups for risk analysis in critical infrastructure such as power plants, 70x speedups for bank-regulated risk modeling (e.g., Solvency II), 40x speedups for circuit simulation using the ngspice simulator, and 60x speedups for event generators in high-energy physics simulations. Ultimately, our product solves enterprise and high-performance pain points around implementation cost, operating cost, and performance.

 

Our hardware modules have been used by commercial institutions and hobbyist developers in drone use cases, for state estimation algorithms such as particle filters, and to retrofit industrial programmable logic controllers (PLCs) for predictive maintenance. For the robotics and industrial automation use cases, our product addresses two complementary pain points: first, being able to retrofit legacy systems by taking advantage of our novel use of a well-known form factor, and second, easier algorithm implementations and faster execution of algorithms ranging from sensor calibration margining (up to 1000× speedup) to particle filters (30× speedup).

3. How does your technology solve this problem?

Our UxHw technology addresses these pain points by providing easier-to-implement software and being orders of magnitude faster than traditional processors, for the use cases we target. Concretely, UxHw uses a combination of four ideas.

 

1. It exposes an architectural abstraction to software that is cleanly separated from the lower-level microarchitectural implementation.
2. The core of microarchitectural innovation is not just brute-force muscle, but a fundamental reduction in computational complexity. We take a problem that scales poorly in software (i.e., improving as the square root of added effort) and make it improve linearly with added resources at the microarchitectural level.
3. Using binary translation as the implementation for the abstraction and restructuring of execution to lower the computational complexity means that, even though we describe the innovation as “microarchitectural”, we can get most of the benefits without actually having hardware.
4. We can get additional speedups if we take the operations inserted by our binary translation (or compile-time transformations) and accelerate them in hardware.

The transformation we apply to programs augments each floating-point operation in a program, adding operations on an in-machine representation for continuous probability distributions. This, in turn, means that as programs execute, in addition to obeying the original program semantics, they also effectively perform arithmetic on distributions, under the hood. As a result, software that would previously have needed iterative sample-based repetition (e.g., Monte Carlo methods) can run in a single pass on UxHw. There are many technical innovations to make this work correctly and fast, including handling correlations. The upshot, however, is massive speedups.

We have a single core technology, UxHw, and our cloud, on-premises, and hardware modules are just different delivery vehicles of this core technology.

4. What application areas are your strongest?

The applications enabled by UxHw are broad, including robotic and physical AI algorithms such as particle filters, portfolio optimization in finance, packing optimization in intralogistics, and both digital and analog simulations in chip design. Our strongest gains are at the moment in three very different areas:

• Robotics and industrial automation, where we have a partnership with Bosch
• Engineering and physics simulations, where we are working with CERN
• Finance (banks and asset managers), where our technology is invaluable for both cost centers like risk modeling required by regulators and profit centers such as digital banking and capital markets

5. What does the competitive landscape look like, and how do you differentiate?

It helps to separate two things: the incumbent we displace, and the emerging field we’re part of.

The incumbents are classical processors from Intel, AMD, and Arm.

The emerging field is what many in the industry are referring to as Probabilistic Computing. Probabilistic computing is the industry’s name for technologies that compute with probability distributions rather than single numbers. Some of these emerging technologies draw on probability theory to represent whole distributions compactly, in a fixed number of bits. Others harness the physics of materials to induce randomness in individual bits. Several Probabilistic Computing startups are moving rapidly to bring products to market.

Our key differentiators from other companies in this space

1. We have a product that is proven to work in practice, whereas many competitors only have technology demonstrations in research papers (or YouTube videos)
2. We’ve demonstrated our product with real applications
3. For real-world applications, our shipping product provides speedups and efficiency improvements of 100-fold or more.
4. Our product is generally available and shipping today, with our cloud platform available in partnership with AWS and our hardware modules available through large global semiconductor distributors including Mouser.

6. How do customers normally engage with your company?

Most customers, even those who purchase our hardware modules from distributors such as Mouser, also start with our free cloud-based Developer Platform (available online at https://get.signaloid.io).

Our developer platform lets developers run applications right away, in a cloud-based compute environment, and has integration with GitHub for hosting codebases. We also provide a command-line tool which developers can use, alongside API keys obtained from the cloud platform, to automate compilation and deployment of compute tasks to our cloud-based compute platform. The command-line tools also give developers an easy way to compile firmware for our hardware modules.

Customers using our hardware modules can purchase them from semiconductor device distributors like Mouser (and soon Digi-Key), and can also purchase them bundled with industrial control automation tools from Bosch Rexroth, our industrial automation partner.

Enterprise customers such as banks or large engineering firms today often deploy some of their workloads on Amazon Web Services (AWS). For such customers, they can purchase a Signaloid AWS machine image (AMI) that they can then deploy on their own cloud AWS instances or on their on-premises AWS variants like AWS Outpost. Enterprise customers who don’t want to deploy their own AWS instances can also deploy their workloads directly over our Signaloid Cloud Compute Engine, which is built on top of AWS and is ISO 27001 and SOC 2 Type II certified.

We have an exciting roadmap of further hardware acceleration for both our cloud and edge hardware. Our recently taped-out ASIC is due to go into a compute cluster deployment in the UK in the coming months and we’ll soon be announcing partnerships with major AI accelerator companies to bring UxHw technology into AI data centers.

It’s important to note that, regardless of the platforms, there is a single core technology: UxHw. The cloud, on-premises, and hardware modules are just different delivery vehicles of that same core technology. Individual developers or large development teams can take existing C or C++ code, compile it using our toolchain, and see results from running their own workload in minutes or hours, rather than days or months. And, because UxHw provides a consistent abstraction across these different platforms, developers can easily move their applications across the deployment vehicles of our technology, with no changes needed.

Share this post via:

• 
• 
• 
•