Most discussions about thermal interface materials center on thermal resistance and junction temperature. And for good reasons! These are clearly very important metrics for any thermal interface material or cooling system. But the reality is that once a thermal interface material is deployed in a data center, it becomes part of the operational workflow needed to keep the data center working as intended. And this means it’s a question of when, not if, some amount of rework will be required. It is therefore crucial that data center operators understand that a TIM that is easy to rework can dramatically reduce cost, risk, and downtime.
What exactly is rework?
Anytime a GPU or CPU is replaced or upgraded or a heat sink or cold plate is removed, a TIM has to be reworked. This means that the existing TIM has to be removed and a new TIM has to be applied, meeting its original factory assembly specifications in order to provide the same cooling benefits as it did on the day it was commissioned. So in practice, any time a GPU or CPU has to be physically inspected or a leaky cold plate needs replacing or an unexpected temperature rise has to be investigated data center technicians have to remove the TIM and install a new TIM when they are done.
Events that trigger rework are therefore not rare edge cases; they are operational inevitabilities. This rework process, however, must be completed without the optimized workflows, controlled environment, and expertise that guided the original manufacturing process. And how difficult or burdensome the rework process is directly affects mean time to repair, repair error rates, and uptime.
The hidden cost of phase change materials
Phase change materials (PCM) are commonly used TIMs due to their relatively low time zero thermal resistance and material cost. They are called “phase change” materials because they transition from a relatively tough and brittle solid to an extremely pliable wax ~20 degrees above room temperature. This means that removing a PCM TIM at room temperature is very difficult and requires aggressive physical scraping, endangering the sensitive and expensive GPU or CPU components the TIM rests on. In practice, PCM rework is therefore usually accomplished by placing entire chip or cooling components into an oven at elevated temperature to significantly soften the PCM, thus easing its removal.
But this process clearly has its own drawbacks: the need for additional CapEx spend on bulky ovens and a place to store them, a longer service window, higher labor costs, potential component damage from heat exposure, etc. I think it’s safe to say that PCM rework is an operational nightmare. And this nightmare adds significant costs, costs that might not show up on a data sheet or during procurement, but costs that a data center operator will bear nonetheless.
Carbice: as easy as peel-and-stick
Carbice is a solid-state TIM that can be easily removed at room temperature. New Carbice pads can be applied simply by peeling the liner protecting the adhesive layer that provides tack during assembly and placing the pad on the CPU/GPU or cooling hardware. No physical scraping, no reheating, no solvents, no fuss. This easy rework and application process enables faster swaps and shorter downtime, more predictable service procedures with lower error rates, and reduced risk to high-value hardware.
Carbice Pad makes rework fast and clean with simple peel-and-stick removal and reinstallation—no reheating, scraping, or process tuning required.
We’ve heard from customers that it can cost nearly $200 per CPU for PCM rework. At the megawatts scale this can mean hundreds of thousands of dollars in rework!
Thinking in terms of total costs
In order to select the lowest cost TIM, it’s essential to move beyond $/unit thinking. To understand which solution is the lowest costs, cost must be evaluated at the system level, encompassing not just material costs, but operational and lifetime costs associated with a particular material. This is the only way to understand the total cost of ownership for a given TIM.
It should be clear at this point that Carbice minimizes operational costs by minimizing service time and complexity and that the operational costs associated with PCM are substantial. We’ve heard from customers that it can cost nearly $200 per CPU for PCM rework. At the megawatts scale this can mean hundreds of thousands of dollars in rework! To minimize total cost of ownership, materials must be chosen that scale operationally, not just thermally. As service frequency increases and hardware value rises, reworkability becomes a primary driver of system-level cost. Carbice was designed to address these operational constraints, so we’re happy to discuss how Carbice can reduce service time and lifetime costs for your next deployment!
