Discrete devices or power modules? This is a question every power electronics engineer faces at some point in their career. I recently had to answer this question for myself and had trouble finding existing literature on the subject, so I thought I’d share my thoughts on the question. There are a number of different tradeoffs to consider -
Power modules have an electrically isolated baseplate with four bolt-holes which makes thermal interfacing much more straightforward
Power modules have a direct bonded copper (DBC) substrate which is used to interface between the semiconductor die and the thermal baseplate. DBC substrate has very good thermal conductivity and provide high voltage isolation
Mounting to a discrete package can be nontrivial. The heatsinking surface may not always be isolated, there may be too little pressure, etc. The point is that there is a higher risk of insufficient mounting here.
There is a higher level of integration with the power module, which means it will be more expensive to replace if a switch is damaged. Whereas with discrete packages, you can individually replace damaged components making repairs much cheaper. However, in many applications, it actually ends up being easier/cheaper for the field service engineer (once you factor in labor costs) to just replace the whole inverter instead of just the power semiconductor.
New semiconductor die technology typically shows up in discretes first. You may have to wait a few more years to get that same die technology in a module. This is because designing a proper power module requires far more effort than putting a single die into a lead frame and molding it into a standard package. We are currently seeing this with Silicon Carbide (SiC) technology, specifically three phase-two level modules. There are only a few three phase SiC modules on the market and they are in high demand. Low Supply + High Demand = Long Lead-times.
Discretes require high current PCB design which can be nontrivial and expensive. This is not realistic for high kW/MW level systems
With power modules, connections in high current path are achieved through bolted busbar connections making it simple to interface with. No specialized PCB services are needed. Additionally, large DC link capacitors typically have busbar connections, so the DC link connection needs to be done with busbars anyways.
Discrete designs inherently have more flexibility and can offer higher power densities if done correctly.
Paralleling TO-247 FETs can help reduce stray inductance if properly done. Special care must be taken to ensure short return paths in the DC link. Imagine having to connect 8 devices in parallel to form a single switch in a 6-pack. You end up with 48 devices that need to be connected to your DC plane. This can pose layout challenges.
In discrete designs, the gate drive circuit can be placed very close to the actual semiconductor, which helps to minimize parasitic inductance. With modules, there is more inductance due to the mechanical and wire-bonded connections.
Based on a report from System Plus Consulting, both modules and discretes are being used in large-production vehicles. Example - Both the drive units in the Tesla Model S and Model III use discrete packages. I can only speculate here, but I imagine this was a cost and supply-chain driven decision. It’s inherently cheaper to make discrete semiconductors (less components and manufacturing processes) and they can be produced at a much higher rate than power modules.
With that said, using power modules is the simpler and more convenient approach to developing medium to high power converters/inverters. You have a higher probability of getting a working product if you use modules. There are just more variables (thermal interfacing, stray inductance, current balancing, etc.) that you need to get right with discrete packages. However, if you are trying to optimize cost and need a steady supply of components for higher volume production, then you may want to further investigate if it’s worth it to use discrete packages.