Chiplet Thermal Co-Design Director — Medical Imaging Compute

The mandate

A reconstruction module places compute, memory and I/O chiplets under a shared lid, but imaging workloads create local heat gradients, package curvature and interface-material movement not predicted by steady-state models. Prototype cold plates control average temperature while pump noise and service access conflict with scanner requirements. The defined problem is to establish an integrated thermal-mechanical design whose lifetime and diagnostic performance survive clinical duty and maintenance.

The deliverables are a workload-to-power model, chiplet thermal map, package warpage correlation, interface-material study, cooling architecture, acoustic and service analysis, transient control strategy, accelerated-reliability matrix, design limits and accepted evidence dossier. Artefacts must bind chiplet revision, package stack, lid, interface material, cold plate, flow and firmware state. Average junction temperature cannot stand in for local cycling and gradient.

Four milestones govern five months: configuration and duty-cycle baseline by 30 October 2026; correlated thermal-mechanical models and concept decisions by 11 December; witnessed transient, reliability and service demonstrations by 29 January 2027; and accepted design envelope, supplier controls and client-led review by 26 February. Milestone evidence includes cold-start, scan bursts, idle recovery and cooling-degradation conditions.

Acceptance requires package teams to reproduce measured curvature, thermal engineers to correlate local temperatures, system leaders to meet acoustics and service limits and reliability to approve the stress rationale. The sponsor returns one consolidated design-defect register within six working days. A cooled benchmark cannot satisfy acceptance if interface pump-out, local memory temperature or clinical-workload transients remain unmeasured.

The client provides package and board models, power telemetry, imaging workloads, thermal samples, interface materials, cold plates, environmental chambers, acoustic limits, service procedures and supplier access. It funds agreed instrumented builds. The director cannot release medical product, change clinical performance claims, select long-term suppliers or approve system safety; authorised client leaders retain those decisions.

Why this is external work

Chiplet package, cooling and scanner teams each optimise different temperature, mechanical, noise and service boundaries. Internal owners also selected the current architecture and need independent challenge before the package is frozen. An external director can connect physical evidence to clinical duty and finish when the client owns explicit, testable design limits.

What you will own

• Translate reconstruction, acquisition, idle, calibration and fault workloads into spatial and transient chiplet power maps.
• Correlate die temperatures, package curvature, lid behaviour, interface pressure, material movement and board constraint.
• Compare air, cold-plate and hybrid cooling through local performance, acoustics, leakage, service and failure recovery.
• Design transient controls that preserve image throughput without hiding inadequate physical heat removal or memory limits.
• Build reliability stresses for thermal cycling, pump-out, delamination, corrosion, flow degradation and maintenance disturbance.
• Define supplier and assembly controls for flatness, material application, clamping, cleanliness, flow and diagnostic sensors.
• Deliver accepted design limits, model correlation, service boundaries, supplier evidence and client-led demonstrations.

Candidate qualifications

• Directed thermal-mechanical co-design for chiplet, 2.5D or other high-power advanced-package systems under production reliability constraints.
• Can evidence an average-temperature pass that concealed local cycling, warpage or interface degradation.
• Understands chiplet power mapping, package mechanics, thermal interfaces, liquid cooling, acoustics and accelerated reliability.
• Has integrated service and clinical or regulated-system constraints into hardware cooling architecture.
• Directed package, board, firmware, cooling and supplier teams through instrumented physical correlation.
• Converts simulation into bounded design rules rather than an unverified high-fidelity model.

Non-negotiables

• Will attend Vancouver, Toronto and Hsinchu design, scanner and supplier sessions.
• Has personally qualified liquid- or high-performance-cooled advanced packages through transient reliability evidence.
• Accepts milestone rejection when average temperatures omit local chiplet, memory or interface conditions.
• Will disclose package, thermal-material, cooling, semiconductor and medical-system relationships before engagement.

1. 49 words maximum. Which package passed average temperature yet failed local mechanical or lifetime evidence?
2. 49 words maximum. How would you represent clinical imaging bursts in a thermal qualification plan?
3. 49 words maximum. What service constraint most often invalidates an otherwise strong cooling architecture?