Mohammad Shahid

Mechanical and Thermal Design Workflow for a 250 W PCIe AI Accelerator Card

Developed a simplified mechanical and thermal design workflow for a high-power PCIe-style AI accelerator card. The project combines first-principles thermal resistance sizing, TIM and airflow sensitivity studies, passive and vapor-chamber tradeoff analysis, forced-air heatsink screening, ANSYS Fluent conjugate heat-transfer CFD, mesh-sensitivity checking, turbulence-model sensitivity checking, concept-level mechanical support screening, and liquid/immersion cooling screening.

The workflow progressed from analytical screening to CFD design iteration. A reduced-height 80 mm × 100 mm × 35 mm heatsink initially failed in an open/bypass domain because a significant portion of the airflow bypassed the fin region. Ducting improved the thermal performance, and extending the ducted heatsink length to 120 mm produced a passing first-pass design.

• Chip heat load: 250 W
• Target maximum chip temperature: 85 °C
• CFD-0 reference case: 80 mm × 100 mm × 50 mm heatsink, 75.97 °C maximum chip temperature
• CFD-1 reduced-height open/bypass case: 91.47 °C, failed target
• CFD-2 ducted 80 mm × 100 mm × 35 mm case: 87.78 °C, slight fail
• CFD-3 ducted 80 mm × 120 mm × 35 mm case: 83.52 °C, passed target
• CFD-4A refined-mesh repeat: 78.11 °C maximum chip temperature, 56.36 Pa pressure drop
• CFD-4B k-omega SST sensitivity: 68.50 °C maximum chip temperature, 83.84 Pa pressure drop
• Mechanical support screening: final heatsink mass approximately 424 g, high support-risk category
• Liquid/immersion screening: compared air, water-like coolant, dielectric liquid, cold plate, and immersion concepts
• Tools: Python, NumPy, Pandas, Matplotlib, ANSYS SpaceClaim, ANSYS Meshing, ANSYS Fluent

Although refined-mesh and k-omega SST sensitivity checks were added, this remains a portfolio-level engineering demonstrator rather than a validated product design. Further mesh-convergence refinement, transition-model assessment, detailed PCB layout checks, mechanical support design, PCB structural FEA, shock/vibration assessment, liquid-cooling pressure-drop modelling, immersion-system CFD, and experimental correlation would be required before product-level validation.

View GitHub project

Coupled Multiphysics Workflow for Thermally Loaded Hardware

Developed sequential simulation workflows linking electromagnetic loss, temperature rise, thermo-mechanical deformation, and response change in thermally loaded high-frequency hardware. The work focused on understanding how coupled physical effects propagate across domains and how model results can inform design sensitivity and performance-risk assessment.

Compact Thermal Screening for Integrated Hardware Systems

Developed first-principles thermal screening models using conduction, radiation, and natural convection to estimate heat-rejection limits in integrated high-power hardware. Simplified predictions were compared with detailed 3D simulations to define model applicability limits and support early-stage design decisions.

Solver Verification and Validation for Plasma-Flow Modelling

Verified and validated an in-house plasma-flow solver using benchmark cases, geometry and boundary-condition checks, mesh refinement, convergence studies, and comparison against analytical, numerical, and experimental references. Cases included Poiseuille flow, de Laval nozzle flow, laminar round jet, and free-burning argon arc simulations.

Industrial Concept Screening with Reduced-Order Thermal Modelling

Developed a compact MATLAB thermal-resistance model for rapid cooling-concept screening using geometry, heat-load definitions, material assumptions, and design constraints. The model was benchmarked against higher-fidelity simulations to identify when simplified modelling was sufficient and when detailed simulation was required.

Unsteady Thermo-Fluid Simulation Around an Oscillating Heated Cylinder

Adapted an in-house finite-difference Fortran solver for unsteady thermo-fluid simulations around a torsionally oscillating heated cylinder with prescribed boundary motion. The work included grid-independence checks, time-step sensitivity studies, wake-dynamics analysis, force response, and heat-transfer trends.

XFEM Fracture Mechanics and Crack Propagation

MSc computational mechanics project using XFEM/Xcrack tools to study crack-tip enrichment, convergence behaviour, stress-intensity-factor extraction, and Brazilian fracture-test response in 2D plane-stress elastic domains.

View report

Fluid-Structure Interaction Benchmark Studies

Numerical study of FSI benchmark problems including imposed airfoil motion, added-mass effects, spring-mounted response, and dam-break impact on rigid and elastic walls. The work included grid, time-step, and modal-sensitivity studies.

View report

POD/PGD Model Reduction and Plane-Stress Elasticity

Implemented POD and PGD reduced-order modelling workflows for transient heat conduction, Poisson problems, and vector-field plane-stress elasticity. The work compared reduced solutions against full-order or exact references and analysed error convergence.

View report