Comprehensive investigation on current imbalance among parallel chips inside MW-scale IGBT power modules

TL;DR

This paper investigates stray-parameter imbalance among six parallel sections inside a 1.7 kV/1 kA IGBT power module and its effect on current sharing. Q3D extraction, PSpice simulation, and double-pulse experiments show that middle sections conduct more current and dissipate more turn-on loss, with the imbalance increasing at higher frequencies.

Key findings

• The six parallel sections have unequal stray parameters: middle sections 2–4 have lower stray inductance and resistance than lateral sections. At 1 kHz, extracted \(L_x\) is 74, 46, 35, 48, 69, and 90 nH for sections 1–6, while \(R_x\) is 342, 225, 179, 213, 264, and 346 µΩ. At 5 MHz, \(L_x\) is 69, 42, 32, 42, 61, and 81 nH, while \(R_x\) rises to 10065, 6401, 4645, 5622, 7457, and 10171 µΩ.
• Stray inductance is almost constant with frequency, but stray resistance increases with frequency due to skin and proximity effects; the resistance imbalance is higher at 5 MHz than at 1 kHz, implying more obvious current imbalance during switching transients.
• PSpice simulation and experiments agree that sections 2–4 conduct more current than the other sections during turn-on. The turn-on loss of section 3 is almost twice that of section 6, and middle sections have higher accumulated energy loss.
• At higher operating frequencies (1 kHz, 5 kHz, and 10 kHz with 50% duty cycle), total power loss increases significantly and the imbalance slightly increases. A second module from a different manufacturer with the same ratings also showed middle sections slightly more stressed than lateral ones.

Methods

• A detailed CAD geometry of a commercial 1.7 kV/1 kA IGBT module was imported into ANSYS/Q3D, which uses FEM and MoM to extract stray resistance and inductance of the six sections from the power terminals to each section at frequencies from 1 kHz to 5 MHz. Simulations were performed separately at each frequency rather than using Q3D’s frequency sweep.
• PSpice simulations used the Q3D-extracted parameters, a lumped-charge IGBT model, an 84 µH load inductance, and 600 V DC to study turn-on current sharing among the six sections.
• Experimental validation used open modules without silicone gel, with a Rogowski coil inserted into each IGBT section. A non-destructive tester rated at 6 kA/1.1 kV was used for double-pulse tests; module A was tested at 700 V and 600 V, module B at the same conditions, and a frequency study was run at 1 kHz, 5 kHz, and 10 kHz with 50% duty cycle.

Limitations

• The paper states that because the PSpice parameters were extracted at the specific IGBT switching frequency (5 MHz), the simulation results only accurately reflect current sharing during the switching transient; for conduction-state current sharing, parameters at a lower frequency should be included.
• The paper states that Q3D simulations should be performed separately at each frequency because the finite-element mesh is optimized only for the simulated frequency, so accuracy could vary at frequencies significantly far from that solution frequency.
• Evident from the text: the study analyzes only the upper leg of the half-bridge module, uses open modules without silicone gel or plastic frame, and validates with modules from only two manufacturers, so the findings may not fully represent packaged or broader population behavior.