Power and control systems are often designed around the moments when the system is pushed to its limits.

“Every system that was ever designed, ever built, ever made, was sized for the worst microsecond of operation, not the average second,” says Martin MacDonald, COO and co-founder of weeteq. “Those peaks and droops are really what dictate the size of systems.”

For weeteq, a Glasgow-based semiconductor startup, that assumption represents an opportunity to rethink how power and control systems are designed. Instead of designing around worst-case conditions, the company’s Ultra Edge® technology uses machine learning to predict transient behavior and respond in real time (sub-millisecond).

The idea has applications across industries that depend on efficient power and control systems, from motor drives and industrial automation to automotive systems and AI data centers. By anticipating electrical transients before they occur, weeteq aims to reduce the excess margin traditionally built into power and control systems, this then represents significantly reduced component stress and energy savings.

The concept traces back to founder Dr. Taner Dosluoglu’s experience in semiconductor power management and a simple question: what if engineers no longer had to design for the worst microsecond?

Applying Mobile Power Lessons to Industrial Systems

Earlier in his career, Dosluoglu worked on power applications for mobile devices during the transition from simple phones to smartphones.

Traditional mobile phones had relatively predictable power requirements. Smartphones introduced a different challenge. At any moment, a device might be streaming video, browsing the web, or switching between workloads, making power demand far less predictable. The industry responded by dynamically adjusting power delivery based on how the device was being used.

“Taner’s background in power efficiency optimization in mobiles has almost been waiting for the emergence of edge AI and tiny ML type applications,” says Martin MacDonald. “This new field of engineering allows us to do incredibly complex tasks incredibly quickly.”

weeteq is applying a similar principle to industrial power and control systems. Instead of designing around worst-case conditions, Ultra Edge uses lightweight machine learning running directly on existing microcontrollers to predict and respond to changing operating conditions in real time.

weeteq’s Ultra Edge® Controller IC adds circuit-level AI capability to any existing power system. (Image courtesy of weeteq)

For weeteq, AI is not the product story on its own. It is a tool for solving a specific engineering problem. “We use AI as a delivery mechanism for the algorithm,” MacDonald says. “We do very precise, very lightweight neural networks to deliver our innovation and application.”

Ultra Edge focuses on the transient events that occur between two steady operating states. By anticipating those events and responding quickly, the system aims to improve performance without requiring additional sensors, cloud infrastructure, or dedicated hardware accelerators.

Powering More Than Motors

As the company explored potential applications, motors became a natural starting point given the amount of global electricity consumed by motor-driven systems. The same approach is now being explored in power electronics, electric vehicles, and AI infrastructure.

More recently, the rapid growth of AI infrastructure has brought power delivery in data centers to the forefront. weeteq believes Ultra Edge DIVA can help improve how power is delivered to increasingly demanding GPUs by responding to transient events in real time. Having proven this out on lab hardware, the company shares that the approach can deliver up to 30% more tokens per watt without requiring changes to the GPU or software stack.

Ultra Edge® delivers real-time circuit-level AI to eliminate peak current overshoots and reduce regulation margins. (Image courtesy of weeteq)

“The key value that Ultra Edge brings is that we can synchronize two power delivery paths so that the chip gets exactly what it needs,” MacDonald explains.

The work has already gained industry recognition, including Best Paper in the AI/HPC category at the OCP EMEA Summit 2026 and participation in the Nissan-led SUITE project focused on next-generation electric vehicle efficiency.

Building the Product in Simulation First

“We had this idea, we had this feeling that we had this product in the making, but we didn’t know what the product was,” MacDonald recalls.

As a semiconductor startup, weeteq needed a way to explore and validate ideas before committing to expensive hardware development. Simulation became central to that process. The team uses MathWorks tools to move from early concepts into a shared engineering environment where engineers can test ideas, evaluate performance, and experiment safely.

“When we’re experimenting, and MathWorks plays a big part in this, we are properly experimenting,” MacDonald says. “We’re expecting things not to work, we’re expecting things to fail. But we can do that on the MathWorks platform, and we get away with it because it’s a simulated environment.”

Using MATLAB and Simulink, the team models applications such as motors and inverters, designs and tests Ultra Edge in simulation, and validates performance before moving to hardware. The engineering team also uses tools such as Simscape Electrical, Statistics and Machine Learning Toolbox, and Global Optimization Toolbox to model electrical systems, tune control strategies, and analyze results. Once designs are proven, generated code can be deployed to target processors or integrated with customer workflows.

As the technology matures, the same workflow helps the team transition from models to physical hardware. Before developing its own silicon, the team worked in an FPGA environment and used MathWorks tools alongside partner development environments to exchange data and validate behavior.

“Without MathWorks, that process doesn’t happen,” MacDonald adds. “It allowed us to really, whilst it felt like a long time, move from the development of the idea to the hardware being built. It would have been significantly longer and more costly/risky if it wasn’t for MathWorks.”

The Road to Proven Hardware

For MacDonald, one of the biggest challenges of building a semiconductor startup is proving the technology before investors, partners, or customers can fully evaluate the opportunity. “Being a semiconductor startup is many degrees more challenging, certainly technically and from a funding perspective, than a software startup,” MacDonald says. “You have to get so far along the line with the innovation and the idea to prove that you’ve got something tangible.”

Simulation and in-kind support helped weeteq make that progress without the resources of a large semiconductor company. “The support that we got from MathWorks and through Silicon Catalyst, with the in-kind support, allowed us to really get to where we are without significant fundraising,” MacDonald says.

Much of weeteq’s work to date has focused on proving the concept through simulation, modeling, and technical validation. The company recently demonstrated the technology in hardware with a data center prototype. The next step is moving beyond demonstration and into real-world deployment.

“Up to now, we can talk a good game and we can say what we’re able to do, again thanks to MathWorks and the simulations that we’ve done,” MacDonald says. “But taking that into products and actually getting that in front of the teams that are building these applications is the key thing.”

weeteq was founded on a simple question: what if engineers no longer had to design around the worst microsecond? Having demonstrated the technology, the company is now focused on bringing it to market.

Learn more about weeteq: https://weeteq.com/

Learn more about MathWorks Startup Program: https://www.mathworks.com/products/startups.html