Why does thermal analysis matter for signoff?
From PDVerse PnR Interview Handbook · pdVerse Mentor Guide
Short Answer
Temperature changes the numbers every other signoff check depends on. Hotter metal has higher resistance, so IR drop grows; EM limits are tied to temperature and allowed current falls as metal heats; and leakage rises with temperature, adding power and more heat. Thermal analysis, run in ICC2 with Kelvin through analyze_thermal (ICC2), shows where the die is hotter than the temperature the other analyses assumed.
Technical Explanation
- Resistance: RedHawk scales metal resistance with temperature through tech-file coefficients, R_eff = R_nominal x [1 + Coeff_RT1 x (T - T0) + Coeff_RT2 x (T - T0)^2]. A hotspot therefore has more drop for the same current than a uniform-temperature run shows.
- EM: limits are rated at a temperature, and lifetime falls quickly as temperature rises. RedHawk sets the EM temperature globally with
TEMPERATURE_EM(RH) or per layer withTEMPERATURES_EM(RH). A strap at 95% of its limit at the rated temperature can be over it in a hotter region. - Leakage rises steeply with temperature. More leakage means more power and more heat, a loop that thermal analysis helps you see before it becomes a package or cooling problem.
- Timing corners assume one junction temperature for the whole die. Regions hotter than that make the corner optimistic locally. At some low-voltage nodes delay can fall with temperature, so the hottest region is not always the slowest; you check rather than assume.
- Kelvin in ICC2 needs
thermal.tech_file(ICC2); the guide states it must be provided and is not generated in memory. Power and metal profiles are optional and generated in memory if not given. The engine defaults to helios, and grid resolution trades accuracy for runtime. - Multi-scenario thermal runs use
set_thermal_scenario(ICC2), where the ITF file is required, andanalyze_thermal -scenarios {S1 S2}(ICC2).report_thermal_qor(ICC2) reports the results and the GUI shows a thermal profile map. - Thermal analysis feeds the others: the temperatures it finds should be used when EM, IR and timing are checked in the hot regions.
- Power and heat are tied to activity, so the thermal result is only as good as the power profile behind it. A power profile from a light workload gives a cool map; for signoff, use the power of the workload that heats the die most, and keep the scenario and its inputs with the report.
Common Mistake
The Trap: Signing off EM at the default rated temperature while the thermal map shows a compute cluster well above it. The EM report says 92% worst, but at the real local temperature the allowed current is lower and several straps are over the limit. Nothing fails on the tester; the damage shows up as rising resistance and field failures months later.
Follow-up Question & Model Response
"If the signoff corners already use a high temperature, why run thermal analysis at all?"
Candidate Model Response: A corner temperature is one number for the whole die, while real temperature varies with where the power is. Thermal analysis tells you whether any region exceeds that number, and by how much. If none does, you have evidence that the corner is safe, which is itself a signoff item. If some region does, you rerun EM and IR there at the real temperature and check the timing paths that pass through it.
Practical Example
Tapeout Scenario: (illustrative) A Kelvin run shows an average die temperature of 78 C and a hotspot of 112 C over a compute cluster, while EM was signed off at 105 C. Rerunning EM with the cluster's layers at 112 C lowers the allowed current, and three VDD straps go from 92% to 108% of their limit. Widening those straps by 20% brings them back to 90%, and the IR drop in the cluster falls by 3 mV as a side effect. The timing paths through the cluster are rechecked with the higher temperature applied, and they keep positive slack, so the only change the thermal run forced was the three wider straps.
PnR Flow Mentor Guide
Master the Physical Design Implementation Flow
Read the complete 8-chapter PnR Flow Mentor Guide free on the web — library setup through placement, clock tree synthesis, routing, chip finishing, hierarchical implementation, and ECO, all the way to stream-out.
Continue practising