Empirical formula for reverse recovery current (Irr) modeling – Request for validation

[fatma1233]

Hi everyone,
I’m currently working on an FPGA-based behavioral model of an IGBT, inspired by the paper:
“An FPGA-Based IGBT Behavioral Model With High Transient Resolution for Real-Time Simulation of Power Electronic Circuits.”
In that context, I came across (or built) the following empirical formula for estimating the reverse recovery current Irr:

Where:

• Il: the load current before turn-off
• di\dt: rate of change of current
• Tj: junction temperature
• a,b: empirical constants (e.g., a = 0.02, b = 0.005 in some examples)

This form seems to reflect how reverse recovery depends on current slope and temperature, but I’m wondering:
Is this formula commonly accepted in behavioral modeling of IGBTs?
Is there a theoretical justification or source for this kind of dependency (e.g., square root and log terms)?
Can we extract such parameters from datasheets (e.g., Infineon IGW40T120) or is curve fitting the only option?
Are there alternative models for Irr you would recommend for real-time FPGA-based implementations?

Thanks in advance for your guidance! Any reference or experience sharing is highly appreciated.

 

[D.A.(Tony)Stewart]

I wonder why any reference does not define the OEM as not all IGBT's or HV power BJT's behave the same.

Although we can expect terms that relate to the Q of each junction or in terms of Rce(sat), C(0V) with the inductance effects of a very low C junction with ultrafast "snappy" diodes that Infineon defines by the crest shape of how soft the recovery is.

I imagine there are different levels of the model with assumptions of the diode model for saturation current and emission coefficient, for the C-B or B-E junction.


The most common model I could find was;

Other sources

https://www.eeweb.com/reverse-recovery-charge-current-and-time/

How does this work for junction temp? in 'C or 'K
I recall seeing log (Tj) at one time which works for 'K , so negative 'C numbers don't work either way.

Did you find any correlation with datasheets in the article?

 

[dick_freebird]

There's three "fit params" but a lot of dependence on
"di/dt" which is a circuit outcome or forcing. Seems
likely to need a lot of data taken, across conditions
(Tj, dI/dt) to regress. Per specific device P/N.