Thermochemical Reliability of Components in Immersion Solutions

Author: Samantha Yates, PhD, Principal Engineer Platform Quality & Reliability, Intel Corporation; OCP Immersion Reliability Committee Lead Collection: 2024 OCP Global Summit - Cooling Environments & Immersion

Summary

Intel's OCP Immersion Reliability Committee lead, Samantha Yates, PhD, presents a five-step thermochemical reliability methodology developed and applied to Intel Ice Lake (3rd-generation Xeon) CPUs submerged in PAO 6 (polyalphaolefin grade 6) dielectric fluid. The methodology covers FMEA (Failure Mode and Effect Analysis) for thermochemical failure modes, component material compatibility, component thermochemical reliability, system thermochemical reliability, and reliability risk assessment. The methodology was applied as foundational qualification work to inform Intel's reliability risk assessments for specific customer immersion deployments.

A central finding is that "Polymer S" — an anonymized CPU package polymer — showed statistically significant material compatibility failures when soaked in PAO 6: ultimate tensile strength (UTS) decreased, loss tangent behavior changed with temperature shift, and weight change from PAO 6 adsorption and swelling were both observed. Despite these component-level compatibility failures being rated High risk, the component thermochemical reliability test showed that junction-to-case thermal resistance after immersion soak was statistically matched to the air-cooled baseline — rated Low risk. The system-level test, using 24 Coyote Pass (Ice Lake) server platforms in a full PAO 6 rack, showed that CPU thermal performance under PAT (Platform Acceptance Test) workload was statistically matched between clean PAO 6 and contamination-saturated PAO 6 — also rated Low risk. The net Reliability Risk for Polymer S in PAO 6 was assessed as Low overall, despite high material-level incompatibilities.

The system contamination study tracked fluid contaminants via UV-Vis (ultraviolet-visible spectroscopy) spectroscopy and identified the main contamination species as phthalates (from cables), solder pastes, glycol ethers, and metal contamination. The contamination signal rose from clean PAO 6 baseline and saturated over time — consistent with an ideal batch reactor model reaching steady state. A key operational detail: the heat sink solder paste required three material substitutions before a formulation compatible with PAO 6 was found, illustrating that material change processes are a practical prerequisite for immersion deployment readiness. The work is positioned as foundational data for Intel's OCP Immersion Reliability Committee, which is actively recruiting participants to standardize thermochemical testing methodologies across the industry.