The Micro-Newton Trap: How Asymmetric Mounting Pressure Causes Core-to-Core Thermal Delta on Delidded Chiplets

By Rizowan Ahmed (@riz1raj)
Senior Technology Analyst | Covering Enterprise IT, Hardware & Emerging Trends

Delidding a flagship multi-die processor and transitioning to direct-to-die cooling is a significant milestone for high-performance systems architects, extreme overclockers, and hardware enthusiasts. Yet, many engineering teams and performance chasers are met with a baffling diagnostic anomaly: after installing custom delidding jigs, direct-die frames, and premium cold plates, one core idles at a low temperature while another thermal-throttles under an identical AVX-512 workload. This is not a failure of the silicon, nor is it a simple uneven spread of your thermal interface material. It is a predictable, mechanical consequence of asymmetric mounting pressure.

When you strip away the Integrated Heat Spreader (IHS), you remove the structural "averaging" layer designed to mask mechanical tolerances. Without this copper buffer, the raw silicon dies are exposed to non-uniform clamping forces. Understanding how asymmetric mounting pressure causes core to core thermal delta on delidded chiplets requires looking past the macro-level assembly and analyzing the mechanical interfaces of modern packaging.

The Illusion of Flatness: Why Chiplets Reject Uniform Contact

To understand why direct-to-die cooling is so unforgiving, we must first dismantle the myth of coplanarity. In a monolithic CPU, a single piece of silicon presents a continuous, uniform plane to the cold plate. In contrast, modern chiplet-based architectures—such as AMD’s Ryzen chiplet processors or Intel’s Core Ultra tile-based processors—consist of multiple distinct dies mounted on an organic substrate. These dies include:

• Core Complex Dies (CCDs) / Compute Tiles: Manufactured on leading-edge, ultra-dense nodes, which run hot due to high power density.
• I/O Dies (IOD) / Base Tiles: Manufactured on legacy, thicker nodes, running at lower thermal densities but occupying a larger physical footprint.
• Structural Silicon / Dummy Tiles: Often placed on the substrate purely to balance mechanical loads and prevent package warping under pressure.

Even though packaging technologies like TSMC’s CoWoS or Intel’s Foveros aim for high precision, the height of these dies relative to one another (coplanarity) can vary. Furthermore, the organic substrate itself is not perfectly rigid; it can bow, warp, and flex under thermal load and mechanical stress.

When the IHS is present, its thick copper structure distributes the downward force of the Independent Loading Mechanism (ILM) across the entire package, absorbing these micro-variations. Once the IHS is removed, the cold plate of your water block must interface directly with these mismatched silicon surfaces. If the cold plate is tilted by even a fraction of a degree due to uneven mounting torque, the physical gap between the cold plate and the silicon varies across the package. This variation directly translates to a severe thermal gradient.

The Mechanics of Asymmetric Pressure and Wedge Gaps

When mounting a direct-to-die water block (such as the Thermal Grizzly Mycro Direct-Die or custom units), the block is typically secured using a four-point spring-loaded screw system. If you tighten these screws unevenly, you introduce a wedge gap across the CPU package.

Consider a dual-CCD processor where CCD 1 is positioned on one side of the package and CCD 2 is on the other. If the mounting bracket is tightened unevenly, the cold plate will tilt. This tilt creates a wedge-shaped profile in the Thermal Interface Material (TIM) layer:

• The High-Pressure Zone: The TIM is squeezed to an ultra-thin layer. Thermal resistance here is incredibly low, resulting in excellent temperatures.
• The Low-Pressure Zone: The gap between the silicon and the copper cold plate expands. Because thermal conductivity is inversely proportional to the thickness of the interface layer, the thermal resistance of this zone spikes.

This mechanical tilt is a primary driver of core-to-core thermal deltas. While a tiny gap difference sounds trivial, in the realm of direct-to-die cooling, it represents the difference between a highly efficient thermal path and a thermal insulator.

The Fluid Dynamics of Galinstan: Liquid Metal Micro-Voiding

The thermal delta problem is compounded when using liquid metal (typically a eutectic alloy of Gallium, Indium, and Tin, known commercially as Galinstan). Liquid metal is favored for direct-to-die applications due to its high thermal conductivity (~73 W/m·K compared to ~10 W/m·K for premium carbon-based pastes). However, Galinstan behaves as a true liquid with high surface tension and low viscosity, making it highly sensitive to pressure gradients.

When asymmetric mounting pressure is applied to a liquid metal interface, it can trigger a phenomenon known as micro-voiding and capillary migration. This process unfolds over several thermal cycles:

1. Capillary Pressure Gradients

According to the Young-Laplace equation, the capillary pressure within a liquid interface is inversely proportional to the distance between the two solid surfaces. In a wedge gap, the capillary pressure is higher in the narrow, high-pressure zone than in the wider, low-pressure zone. This pressure differential forces the liquid metal to migrate away from the high-pressure area and pool in the low-pressure area.

2. Dewetting and Micro-Void Formation

As the liquid metal is squeezed out of the high-pressure zone, the film thickness in certain regions drops below a critical threshold. The liquid metal film ruptures, causing it to bead up due to its high surface tension. This process, known as dewetting, leaves behind microscopic dry patches or "micro-voids" where there is no physical contact between the silicon and the cold plate. These voids act as microscopic barriers, blocking heat transfer.

3. Thermal Runaway of Isolated Cores

Because the silicon dies are divided into individual high-performance cores, a micro-void sitting directly over a single core’s execution engine can cause that specific core to quickly spike to its thermal limit under load. Meanwhile, adjacent cores situated under a healthy, continuous film of liquid metal remain cool. This is why you observe a significant delta between cores on the same physical die.

Diagnostic Methodology: Identifying the Delta

How do you differentiate between a poor liquid metal application and genuine asymmetric mounting pressure? The diagnostic process requires precision monitoring tools paired with structured, transient thermal testing.

• The Transient Spike Test: Apply a sudden, heavy workload from a cold idle. If the temperature of the hot cores spikes rapidly in less than one second, it indicates a physical void or zero-contact zone over those cores. A gradual rise indicates a heat dissipation bottleneck, not a mounting issue.
• The Core-to-Core Delta Threshold: Under a uniform, all-core workload, a healthy direct-to-die mount on a chiplet processor should exhibit a minimal core-to-core delta. A large delta is a definitive indicator of asymmetric pressure or localized TIM failure.
• Per-CCD Analysis: On dual-CCD processors, compare the average temperature of CCD 1 against CCD 2. If one CCD is consistently running significantly cooler than the other despite identical power draw, your cold plate is tilted, applying excessive pressure on one side and lifting off the other.

The Precision Remediation Protocol

Resolving asymmetric mounting pressure requires a calibrated, systematic approach to mechanical assembly.

1. Utilize a Calibrated Torque Driver

To eliminate asymmetric pressure, use a calibrated torque limiter (such as a preset 0.2 N·m torque key). Tighten the mounting screws in a strict crisscross (X) pattern, turning each screw by no more than half a turn at a time until the torque limiter clicks.

2. Verify Frame and Substrate Alignment

Direct-to-die contact frames must sit flush with the motherboard PCB. Ensure that the motherboard backplate is not warped and that no surface-mounted devices (SMDs) or capacitors are interfering with the frame. Even a tiny piece of debris under the frame can prevent it from sitting flat, translating directly to an uneven mount.

3. Consider Phase-Change Materials (PCM) as an Alternative

If your mounting hardware cannot guarantee perfect coplanarity, liquid metal may not be the optimal choice. Transitioning to a high-performance phase-change material, such as Honeywell PTM7950, can mitigate minor mounting imbalances. PTM7950 solidifies at room temperature and melts at approximately 45°C. In its liquid phase, it exhibits excellent wetting properties, but unlike liquid metal, its polymer matrix resists pump-out and micro-voiding under uneven pressure gradients.

---