How Ventiva’s Ionic Cooling Solves the Memory-Placement Problem Fans Can’tBy Brian Cumpston, VP of Engineering, Ventiva

Running AI models locally on a laptop comes down to a hardware bottleneck most people never think about: how fast memory can be transferred to and from the processor. Solving that bottleneck means placing more memory chips closer to the SoC than laptop architecture has ever required before, and that real estate is already spoken for by the fan.

Moor Insights & Strategy’s recent analysis of AI-ready PC design puts a number on this: a 256-bit memory interface requires eight length-matched, shielded 32-bit channels, with trace lengths under 25mm from the SoC. That’s not a design guideline – it’s a signal integrity requirement. Miss it, and the bus doesn’t run at spec. That 25mm radius is, in nearly every thin-and-light laptop shipping today, exactly where the fan and heat pipe assembly live.

This blog is about what happens when you remove the fan from that 25mm radius, and why ionic cooling is the architecture that actually fits the geometry that the Moor Insights research describes.

Why You Can’t Just Shrink the Fan

The instinct is to make the fan smaller or move it, but neither option works.

A centrifugal fan’s airflow scales with impeller diameter. If you shrink the fan, you either lose airflow or spin faster to compensate, and higher RPM means more noise and wear which is a shorter path to failure. With a sub-28dBA acoustic target, that trade isn’t available. So, the industry has done the opposite: fans have grown to keep RPM down, consuming more board area, not less.

Relocating the fan doesn’t solve the problem either. Pushing the fan toward the chassis edge to clear space near the CPU grows the laptop’s footprint. This is especially true in more compact systems like 14” laptops. The problem just can’t be solved with bigger fans.

What Ionic Cooling Does Differently

Ionic cooling (aka electrohydrodynamic (EHD) airflow) doesn’t move air by spinning anything. A current through a thin emitter wire ionizes air molecules, which accelerate toward an oppositely charged collector; collisions with neutral air along the way drive a bulk outflow. So, there’s no need for an impeller, blades, and a rotating mass.

This has two geometric consequences:

• Flow is side-in, side-out, which eliminates the air plenum and the Z-height it costs.
• The device is rectilinear, not circular: a 60mm fan claims roughly 60mm x 60mm of board area because its impeller is a disc. An ionic cooling module with the same airflow-direction length might be 60mm x 20mm. Same length where length matters but a fraction of the width – and that’s the footprint that clears out of the processor’s 25mm radius.

The trade-off is real: ionic cooling devices operate at a lower total pressure head than a fan, typically 10-15 Pa versus 60-70 Pa for a laptop blower. This means designers have to be more deliberate about inlet and outlet impedance, but it doesn’t block a working design. (Our laptop reference design demonstrates this.) It just means the ducting and vent geometry around the module matter more, which is exactly the type of system-level design work we do with customers.

The Board, Before and After

Picture a 28W thin-and-light notebook: two circular fan cutouts flanking the CPU and memory routed around whatever board space is left, often with longer traces or added layers to compensate for the detour.

Now replace those cutouts and take a Zoned Cooling design approach: ionic modules can be placed in a shallow rectangular “cooling” zone along one edge, with targeted SoC and skin-temperature airflow running in parallel paths instead of sharing one channel. The board becomes a contiguous rectangle instead of a shape interrupted by two circular voids.

What goes into that reclaimed space is an engineering choice, and the priority order is straightforward. Memory first: higher local AI inference performance requires higher memory bandwidth, which requires more memory packages closer to the CPU. Then additional M.2 storage or wireless card space, since a contiguous board accommodates full-length modules a fan cutout would otherwise interrupt. Battery capacity is the third lever, contingent on how much volume is freed front-to-back in the chassis, but it’s a real option once the thermal solution stops competing with the pack for the same real estate.

Why the Timing Matches Moor Insights’ Forecast

Moor Insights’ white paper puts today’s typical thin-and-light laptop bandwidth at roughly 150 GB/s, capped by 128-bit LPDDR5, which is well short of what frontier-quality local AI inference needs. The path forward is a wider bus and a shift to LPDDR6, which turns the 25mm constraint into the default constraint every OEM will design against.

Our laptop reference design – a 14″ notebook at 28W SoC power, sub-16mm chassis thickness – sits inside that range because that’s where the geometric pressure is most acute today. As bus widths grow and TDPs climb, the same principle holds: airflow generation that doesn’t require a circular, plenum-fed footprint is the only way to make additional memory real estate available at all.

The Floor Plan Was Never Fixed

The industry has spent three decades treating the fan’s footprint as a constant. The issue was never solved – it simply remained unquestioned. Solving this system design constraint, not the airflow problem, is the key to unlocking on-device AI inference in a laptop chassis. That’s why we built ionic cooling around the geometry the board needs, not around the shape of the fan it replaces.

For the analysis behind this shift, read the white paper: How AI is Reshaping Modern PC Designs.