2.5D / CoWoS Package Thermal & Warpage Simulator

Model whether an advanced package survives sign-off — then run it: the simulation executes on the ChipFoundryServices distributed compute pool. A high-power AI accelerator is a stack of silicon on an organic substrate, and two physical failures gate whether it ships. Thermally, junction heat escapes through a series of resistances — the die-to-lid TIM, the copper lid, the lid-to-heatsink TIM and the heatsink — that add to θ_JA, giving Tj = T_ambient + P·θ_JA; if Tj exceeds Tj_max the part throttles. Mechanically, the silicon interposer (CTE ≈ 2.6 ppm/°C) bonded to an organic substrate (CTE ≈ 17 ppm/°C) bends like a bimetallic strip on cooling from the mold-cure temperature; Timoshenko curvature times the package span squared gives the coplanarity warpage, which stiffener rings and balanced mold caps knock down. Too much warpage opens C4/BGA joints. The package passes only if it clears both gates. Reduced-order educational model. See also the chiplet / CoWoS, HBM bandwidth, network-on-chip, electromigration, reliability / aging, die thermal, power-delivery network and interconnect RC simulators and the compute-pool status.

AI accelerator · CoWoS HPC GPU Mobile SoC
Thermal — heat path to ambient
Warpage — CTE-mismatch bending
Junction temperature Tj vs package power — Tj rises linearly at slope θ_JA off the ambient floor. It crosses the dashed red Tj_max line where the part throttles; the marker (●) is your operating point. Steeper slope ⇒ a worse thermal stack
Left: coplanarity warpage vs stress-free temperature — a bigger drop to room temperature bends the bimetal more; it crosses the dashed red spec where the package fails coplanarity (marker ● = your mold-cure temp). Right: the θ_JA resistance stack — each layer as a share of the junction-to-ambient path; the amber bar is the dominant resistor to attack first
Developer API — same simulation over HTTP (load-balanced across the pool):
curl -X POST https://www.chipfoundryservices.com/edge/warpage \
  -H "Content-Type: application/json" \
  -d '{"total_power_w":350,"die_area_mm2":600,"tim1_thickness_um":50,"tim1_conductivity_w_mk":5,
       "tim2_thickness_um":200,"sink_resistance_c_w":0.08,"ambient_c":35,"tj_max_c":105,
       "substrate_size_mm":55,"interposer_thickness_um":100,"substrate_cte_ppm":17,
       "stress_free_temp_c":180,"warpage_balance_factor":0.82,"warpage_spec_um":150}'
Returns JSON with outputs (theta_ja_c_w, r_tim1_c_w, r_lid_c_w, r_tim2_c_w, r_sink_c_w, junction_temp_c, thermal_headroom_c, meets_thermal, dominant_thermal_resistor, delta_cte_ppm, delta_temp_c, curvature_1_m, raw_warpage_um, net_warpage_um, warpage_headroom_percent, meets_warpage, warpage_shape, meets_budget, verdict), the full profile (48-point tj_vs_power sweep [power_w, junction_temp_c] and 48-point warp_vs_temp sweep [stress_free_temp_c, net_warpage_um]), the serving node, and compute_ms. Endpoint aliases /edge/packagewarp, /edge/packagethermal, /edge/coplanarity, /edge/ctemismatch, /edge/thetaja, /edge/junctiontemp.