Model where a processor’s watts go — then run it: the simulation executes on
the ChipFoundryServices distributed compute pool. Every switching transistor burns dynamic power
≈ activity × C × V² × f — quadratic in voltage, linear in clock —
and the wires add interconnect power on top. But the design also leaks: static leakage rises
with temperature and doubles roughly every 15 °C, so leakage heats the die, which raises
leakage further. The node solves that junction-temperature fixed point
(Tj = ambient + θJA × Ptotal), checks the design against its
TDP and Tjmax, flags thermal runaway when the leakage–temperature loop diverges, and
finds the headline number: the thermal-limited maximum clock the cooling solution can sustain. These
are the same dynamic-vs-leakage and voltage-vs-frequency trade-offs that set an AI accelerator’s power
envelope. Reduced-order educational model. See also the
training-memory, KV-cache, HBM bandwidth,
systolic array, transistor I-V,
thermal, interconnect RC,
die-yield, 6T SRAM, CMP planarization and
lithography simulators and the
compute-pool status.
curl -X POST https://www.chipfoundryservices.com/edge/power \
-H "Content-Type: application/json" \
-d '{"transistor_count_b":20,"core_voltage_v":0.85,"clock_ghz":2.0,
"activity_factor":0.15,"cap_per_transistor_ff":0.10,"interconnect_fraction":0.4,
"die_area_mm2":600,"leakage_density_w_mm2":0.015,
"thermal_resistance_c_w":0.06,"tdp_w":700}'
Returns JSON with outputs (dynamic_power_w, logic_dynamic_w, interconnect_power_w,
leakage_power_w, total_power_w, junction_temp_c, tjmax_c, tdp_w, power_density_w_mm2,
leakage_fraction_percent, within_tdp, within_tjmax, thermal_runaway, max_clock_ghz,
energy_per_cycle_nj, gops_per_w, verdict), the full profile (48-point
power_vs_freq and power_vs_volt sweeps, plus dynamic_w,
leakage_w, tdp_w, total_w, clock_ghz,
core_voltage_v, junction_temp_c, within_tdp), the serving
node, and compute_ms. Endpoint aliases /edge/powerbudget,
/edge/tdp, /edge/dvfs, /edge/leakage.