Thermal throttling in compact devices is a performance limit triggered when a mobile processor's sustained heat output exceeds what passive cooling can dissipate, forcing a reduction in clock speed. Edge AI is accelerating this problem fast — not because compact devices are approaching datacenter-level power dissipation (they are not), but because the density of compute relative to available cooling surface has never been higher. Bridging this gap requires borrowing datacenter cooling principles — direct-to-chip liquid flow and dynamic flow control — and re-engineering them for a 7mm form factor that runs on a battery.
Key takeaways
- Edge AI brings datacenter-level workloads to smartphones, laptops, and smart glasses, creating new thermal demands.
- Traditional mechanical pumps cannot fit within the 1.2mm stack heights required for compact device cooling.
- Piezo-actuated microfluidics enable real-time flow rate modulation to match dynamic AI compute spikes.
- CapDrive™ energy recovery reduces the electrical overhead of active cooling in battery-powered devices.
- Passive cooling cannot sustain the heat loads of local AI inference in mobile processors.
Microsoft recently published a great piece on why microfluidics are the future of liquid cooling for AI chips. They rightly point out that to handle massive Neural Processing Units (NPUs), the industry has to get the cooling fluid directly to the silicon.
We completely agree: liquid cooling is the definitive solution for AI. But here is the catch—AI isn't staying in the datacenter.
Edge AI is bringing heavy, datacenter-level workloads directly into smartphones, laptops, and smart glasses. If consumer devices are going to run these intensive local AI models, they need serious Edge AI thermal management.
This creates a massive engineering disconnect. The current liquid cooling technology used for server racks is simply not built for consumer devices. You can build the most advanced microchannels in the world to cool a mobile chip, but you still need a way to push fluid through them without ruining the form factor.
I asked Cedric Leclerc, our liquid cooling expert at Boréas Technologies, for his take on the shift toward micropump liquid cooling. Here is why he argues that piezo-actuation is the only practical way to scale this technology down for consumer hardware.
1. Mechanical pumps simply don't fit
The biggest issue with direct-to-chip cooling in consumer devices is stack height. You cannot cram standard pumps into a 7mm thick smartphone or a pair of AR glasses.
When designing high TDP cooling compact devices, you urgently need reliable alternatives to passive cooling electronics. Cedric explains why the fundamental mechanics of traditional pumps are already obsolete:
"Etching channels in silicon is only half the battle; you need a heart to move the coolant. Traditional mechanical pumps are fossils that cannot fit within the tight 1.2mm stack heights required for direct-to-chip or side-mounted cooling. Piezo-actuated microfluidics are the only logical path for compact Edge AI devices, providing the necessary pressure heads in a solid-state form factor to move heat to cooler zones within the device without the mechanical failure points of bearings or seals."
You can't rely on moving parts, seals, and bearings in ultra-compact consumer tech. You need solid-state reliability that actually fits the form factor.
2. AI compute spikes. Your cooling should too.
Whether on a server or a smartphone, AI workloads don't run at a constant speed. They come in heavy, sudden bursts. If your cooling pump only has a basic "on/off" setting, the system will lag behind the thermal spike.
To achieve true high heat flux cooling small form factor, the system must be dynamic. Cedric points out how the BOS1921 piezo driver fixes this lag:
"The BOS1921 acts as the high-resolution 'throttle' for these systems. Because AI workloads are inherently spiky, cooling cannot be binary. Our driver allows for on-the-fly modulation of both frequency and amplitude, giving engineers the ability to finely tune flow rates in real-time. This ensures that the coolant velocity is always matching to the instantaneous heat load of the SoC , preventing thermal lag during sudden compute bursts."
Real-time control allows you to match fluid velocity to the dynamic power load of the mobile chip at any given millisecond. That stops thermal throttling before it ruins the user experience.
3. Active cooling shouldn't kill the battery
In a datacenter, efficiency is about lowering the power bill. In consumer tech, it's about battery life. Every watt spent running the cooling system is a watt stolen from the device's runtime.
If you want effective compact active thermal management, the driver cannot generate its own heat. Cedric explains how our piezo architecture flips this power dynamic:
"Efficiency is the ultimate constraint in battery-powered devices. It is counterproductive to cool a mobile processor using a pump driver that adds its own significant thermal load to the system. By utilizing CapDrive™ technology to recover energy during the piezo's discharge cycle, we ensure the power budget is spent on compute, not on the cooling overhead. We are essentially making the pumping mechanism electrically 'invisible' to the overall system efficiency."
Energy recovery matters just as much in a smartphone as it does in a server rack. Your thermal management system has to be electrically invisible.
The Next Step in Active Micro-Cooling
If you are designing thermal solutions for the next generation of compact consumer devices, legacy mechanics are hitting a wall. Miniaturized active cooling systems are the only way to get serious heat transfer with a minimal power draw.
Explore how we are shrinking micro liquid cooling for electronics to fit the future of Edge AI at Boréas Technologies.
👉 Discover our Micropump Liquid Cooling Solutions
4. Frequently asked questions
What is thermal throttling and why does it matter for AI performance on smartphones?
Thermal throttling is a built-in CPU/NPU protection mechanism that automatically reduces clock speed when chip temperature exceeds a safe threshold. On smartphones running local AI models, this means sustained inference tasks — image recognition, language processing, real-time translation — slow down mid-task, degrading user experience without any visible warning.
Why can't traditional liquid cooling systems be miniaturized for smartphones or AR glasses?
Traditional liquid cooling relies on mechanical pumps with rotating parts, bearings, and seals — components that cannot fit within the 1.2mm stack heights typical of compact consumer devices. They also introduce failure points that are unacceptable in sealed, drop-resistant consumer hardware. Solid-state alternatives are required.
What is a piezo-actuated micropump and how does it differ from a mechanical pump?
A piezo-actuated micropump uses the deformation of a piezoelectric element — rather than a rotating motor — to move fluid through microchannels. It has no bearings, no seals, and no moving mechanical parts, making it small enough for 1.2mm stack heights and reliable enough for long-term consumer device use.
Does running an active liquid cooling pump drain battery life significantly?
It depends on the driver architecture. Systems using CapDrive™ technology recover energy during the piezo element's discharge cycle, reducing net power consumption substantially. The goal is to make the pump electrically invisible to the system power budget — so battery runtime is determined by compute load, not cooling overhead.
What makes piezo-actuated microfluidics the right fit for Edge AI devices specifically?
Edge AI devices combine three constraints that rule out most cooling approaches: strict size limits (thin form factors), variable heat loads (bursty AI inference), and battery dependency (no external power). Piezo-actuated microfluidics address all three — they are compact, dynamically controllable, and efficient enough to run off a mobile battery.
6. Related reading
- Micropump liquid cooling solutions for compact electronics — Overview of Boréas's piezo-actuated micropump platform for direct-to-chip and side-mounted cooling applications.
- BOS1931 ultra-low-power piezo driver — Product page for the driver IC used to control flow rate in real-time cooling systems.
- More articles on thermal management and cooling — All Boréas blog content tagged under cooling, including Edge AI thermal challenges.
- BOS1921 development kit — Evaluation hardware for engineers prototyping piezo-actuated micropump systems.
About the author
Marc-André Morin is a marketing specialist at Boréas Technologies, where he leads go-to-market for CapDrive® haptic solutions across wearables and consumer electronics. LinkedIn
Leave a comment