Steve Conner, Senior Principal System Architect

The semiconductor industry has been chasing Moore’s Law for decades, but we’ve hit a wall with power efficiency in control systems. Every engineer I know is wrestling with the same fundamental problem: traditional control loops can’t keep up with the non-linear realities of modern power electronics.

I’ve spent years debugging systems where theoretical models work perfectly in simulation but fall apart when you introduce real-world variables—temperature drift, component aging, load variations. The standard approach is to add safety margins, but that’s just engineering around the problem, not solving it.

What’s fascinating about Ultra Edge® is how it sidesteps the computational complexity that’s plagued embedded AI. Instead of trying to run massive neural networks on constrained hardware, we’re using lightweight ML models that learn the specific non-linearities of each system. It’s elegant—the AI doesn’t need to understand motors in general, just this specific motor under these conditions.

The circuit-level approach reminds me of early FPGA work—getting closer to the hardware to unlock performance you can’t achieve at higher abstraction levels. When I see those smooth torque transitions with zero overshoot, it’s not just better specs on paper. It’s fundamentally different control physics.

We’re finally closing the loop between real-time sensing and real-time correction at the speeds these systems actually operate. That’s been the holy grail for power engineers since I started in this field.

Anahita Kanade – Machine Learning Engineer

Most edge AI applications I’ve worked on follow similar patterns: collection of data through sensors, training of neural networks in the cloud and deployment of static models onto the device. But power electronics don’t behave like image recognition or natural language processing applications—the physics are deterministic, but the interactions are incredibly complex.

The breakthrough with Ultra Edge® was realizing we didn’t need historical data or offline training of large models. These systems generate their own training data in real-time through normal operation. Instead of trying to predict what might happen, we learn what actually happens in a specific circuit configuration.

Traditional control theory assumes linear systems, but any power engineer will tell you that’s fiction. Motor inductance changes with current, temperature affects resistance, and component aging shifts everything. Classical AI would try to model all these variables explicitly, but we’re taking a different approach—let the system teach itself.

The real challenge was inference latency. Cloud-based ML is measured in milliseconds, but we needed microsecond response times. Our models had to be incredibly lightweight yet robust enough to handle the continuous adaptation these systems require.

What’s remarkable is watching the system self-learn and self-heal. Initially, it follows traditional PID behaviour, but with the implementation of Ultra Edge® it starts discovering more efficient control paths that no human engineer programmed. It’s genuinely learning optimal control strategies through direct interaction with the operational data available at a circuit level.

Lewis Gallagher – Sales and Marketing Manager

Power optimisation applies universally across sectors. Manufacturing facilities using servo drives for precision machining, automotive manufacturers seeking EV powertrain enhancement, and wind farms maintaining grid stability. These sectors all benefit from the same underlying improvements – more efficient power conversion and reduced equipment stress.

Our retrofit approach eliminates adoption barriers. Whether integrating with existing controllers through companion hardware or deploying embedded software on current MCUs, customers see immediate performance gains without operational disruption.

Different industries emphasise different benefits from the same core technology. Automotive applications focus on component downsizing and safety compliance, enabling smaller windings and housings while meeting ISO standards. Industrial automation values extended equipment life and improved throughput through reduced system stress. Energy systems prioritise grid stability and efficiency optimisation, supporting renewable integration while maintaining reliability.

The fundamental power electronics improvements deliver measurable value across all applications. ROI becomes evident through operational metrics customers can verify – lower energy costs, extended component lifetimes, and improved system performance