In large systems, one can replace individual failed components, where an integrated solutions can be costly to maintain (e.g. 10kW upwards). For smaller systems (TO220/247 etc) one can use whatever you like. as the expense in replacing a blown unit is minimal. In general, one would optimize the design to minimize the cost of production, which is volume dependent.

Depends on the system you are building. Discretes are the choice if the power circuit is on a PCB (for relatively lower power systems) while modules are selected for the relatively larger power wired systems. Power interconnections between devices is convenient and minimum using modules. Modules mostly come in a package of two devices connected in series, till the power rating is really large, when it becomes single device per module. This gives an advantage sometimes, like one device plus one freewheel diode for a chopper within a single package. Modules are always electrically isolated from their heat sinking surface implying that heat sinks don't need electrical isolation. Discretes may not always be isolated, thus may need isolation for the heatsink from the metal body or isolation for the device from the heatsink. Fixing of non-isolated discrete devices to heatsink with insulation between the two also needs special care as the fixing screw (now electrically connected to heatsink) can become electrically connected to the metal tab of the package, which is one terminal of the device. Criterion of lower cost does often make discretes the preferred ones, notwithstanding associated difficulties, even for larger power applications with several devices used in parallel.

I think it depends on what are you designing, but perhaps the most cost efficient should be the best, having account in quality of course, perhaps dimensions are not the most important issue sometimes. And different components are not completely the same, that´s something that you should have in account.

Thank you, gentlemen, for your valued feedback. All of you pointed out that the cost of production and ownership and even repair is an important issue. You further differentiate discretes and modules by the power level the semiconductors need to handle and see modules to take over if multiple devices need to be put in parallel.

An very important aspect is also the heat transfer capability that is significantly larger for modules since the used DCB-Substrate enhances heat spreading while providing a very good electrical insulation.

The point which is missing yet, is inductance: TO247 is simply not designed to provide a minimum inductance. In contrast, modules may provide very low inductances that make the use of fast switching devices (Si Coolmos, SiC JFET, BJT and MOSFET) economically feasible.

Is there somebody who already made experiences with SiC power modules? How is their performance?

I would say that modules in general have much higher inductance than for instance parallel-connected TO247 packages. The reason to this is that all current is concentrated in one lead in the module. This holds for all connections. If you use several discrete devices in parallel, the inductances of each path are not perfectly coupled, and to some extent the fields even cancel out. Typically, you could, therefore, achieve a lower total inductance if you use discrete devices in combination with a smart circuit layout. The Poseidon module from Infineon has, however, shown that it is possible to make a power module with low inductance. Having said this, I also want to point out that the inductance of the gate loop may be at least as important as the inductance in the main circuit. This depends on your application and on your desired switching speed.

Prof. Nee, I totally agree - beside the power loop inductance, gate drive inductance is definitely a major issue. Many module manufacturers simply leave some small copper islands on the DCBs and make jumpin' wire connections for the gate driveconnections. What's seen often is that a gate return path is provided which does not share the source / emitter wire bonds with the power path. For high current modules this is highly advisable in order to prevent source/ emitter current feedback. But how can gate inductance be dealt with? Do you think that gate drives integrated into modules will soon find their share in the market?

I am sure that "intelligent" power modules wind be more common in the future. Semikron has been working in this direction for many years, but where we will need it most might be for SiC devices, because we want to switch them faster. In the future high-temperature solutions for this purpose must also be provided because if the power device gets hot (SiC can operate at very high temperatures) and the gate driver is close, it will also get hot. One solution is to make the gate driver in SiC as well. This has been investigated by Prof. Carl-Mikael Zetterling at KTH Royal Institute of Technology in Sweden. Also the company Cissoid has been working on high-temperature drivers, but not in SiC.

Prof. Nee,

thank you very much for your interest. Honestly, I believe that "high temperature" is a buzz word of todays power electronics magazines and research proposals. Very high temperature is very undefined, in fact. Actually, there are not so many applications where die temperatures really need to exceed the Si limit of 175°C. And if, it's just the die itself that really goes hot - but it's surroundings may stay rather cool (20-50K lower) allowing for operation of Si dies with comparable low power dissipation. From my perspective, the dT occurs right inside and just below the small SiC dies (due to heat flux density) - but several mm away, a gate drive die would enjoy it's life.

We all should keep in mind that ceramic packages - other than ceramic substrates as DCBs - are extremely expensive, especially if they are made of multiple parts. Thus, as people mean high temperature, either they are referring to applications where cost is secondary or they imply that the actual package, i.e. the plastic frame and the PCB/ busbar mounted on top of that, stay as they are - not rated for continuously >>200 °C ... .

Holger, thanks a lot for your valuable remarks! As far as I am aware, the second source is provided for many IPEMs and "special" module structures and topologies. At least after a while, the original producer gives away licenses to his competitors in order to allow the market to grow - and with it his market share .... .

Holger,

I totally agree with you regarding the problem of "mechanical concentration" in the vicinity of the module. In fact, power input, power output, gate drive, DC-link capacitors, output filters, heat removal (heat sink / liquid cooler) and mechanical structures need to be close to the module. Depending on the operating frequency, the power and the cooling strategy, one ore two of the mentioned subsystems may tolerate a larger distance to the module. But in the end, you don't get a system shrunk together like a black hole. For instance, double-sided cooling may improve the thermal path (but never more than to double performance) but on the other hand severely hinders access to the semiconductors leaving only the x-y plane for interconnection. I personally believe that this strategy needs to be put under question if low electrical parasitics are in focus.

Dear Forum,

a lot has been said about parasitics here and I like to hint out, that these are not limited to the power semiconductor in use.

Paralleling TO247 does help in reducing the stray inductance of the single switch if properly done - agreed.

But afterwards, care has to be taken, because the larger distances to connect a DC-side to the paralleled devices will eventually eat up the gain.

Imagine to parallel 8 or 10 devices to form a single switch in a 6-pack.

Ending up with 48 to 60 devices to be connected to a DC-plane will get you into trouble.

In such cases, modules - or even paralleled small modules - may improve the situation.

However, from mounting, assembly and cost situation, the module most likely outperforms the discretes.

A contradicting example however is the inverter in Tesla's Model S that is built on discretes only :-)

Well - the exception to the rule, I'd think.

An additional question should be, where the application is installed later.

In Europe, maintenance is so expansive that an inverter

One more addition in regard to "Power Electronics".

Without adding a power range, Power Electronics can be anything from mW to MW.

In case you exceed the high kilowatt range, discretes may not be even considered.

Try handling the MW of an offshore windmill using discrete packages :-)

The same occurs if you would try to use modules in switch mode power supplies to charge a mobile phone...

Holger, thank you for contributing these different views towards modules. I also recognized the fact that cutting-edge designs often run with discretes. Besides the flexibilty there may also be the reason that novel semiconductor devices often show up as discrete first. If you want modules, you have to wait further years.

My personal believe is that modules will succeed if they not only do better in thermal matters but also in the electromagnetic field. In the long run, technologies as SKIN may change the rules modules are designed and improve their performance leaving the best discrete design behind ...

Dear Holger,

please keep in mind that any part you integrate into a component takes away the degrees of freedom in the customer's design.

Especially larger companies with dedicated development departments tend to build entities like drivers or busbars to achieve special features, making their design unique.

Taking this opportunity away would lead to an empty business case.

For new chip technologies in modules, the drawback is inherently clear:

Designing a proper power module is far more effort than putting a single die or two to a lead frame and mold it into a standard package.

For sure you'll first find customers to use discretes and once have numbers that allow for building a module, you migrate the new technology to module designs.

Dear Holger,

though I agree to most of what you mentioned, please forgive me if I use your example to substantiate my point:

Integrating "some driver transistors" might get you in trouble as most likely this particular transistors are not qualified parts with the rest of the world. Not every manufacturer of transistors is a welcome supplier around the world.

Additionally, the transistors would implement a certain structure of gate drivers.

So what if - especially to more accurately control the next generation of high-speed switches - the world turns away from voltage source drivers and makes use of gate current control?

The design would be rendered useless, an investment of a few millions down the drain.

Though integration is tempting from engineering point of view, it will always reduce the market you can serve.

Compare the number of so-called IPMs sold in inverters 20kW up to 50kW to the number of typical power modules and you'll see what I mean.

TO220, TO247 are not always easy to mount to the heat sink. There is always the risk of bad mouinting (lack of pressure, or to much, lack of thermal compund, statics, insulation problems, etc.). In my long working life I have seen many problems caused by bad mounting of these discrete devices. 

What I prefer is to use SMT power devices on aluminum substrate.  I have used this aproach for AC/DC converters (full bridge phase shift) up to 1500 W. But it depends on what is available in the market. There is no need to reinvent the wheel. For exemple  for  LCC converters up to 400 W  i have used  an excellent Fairchild module; for three-phase inverters the simplest solution is a module, if possible incorporating the gate drivers. Anything but not TO220, TO247.

Dear Francesc, dear gentlemen

thank you for allowing us to share your experience with us. In conclusion, we find that modules offer

I think, time has come where engineers call for solutions in between: medium voltage, medium current modules with partially included DC-link capacitor and easy-to-use integrated gate drive: no HF-loops allowed to leave the module.

I forgot to inform of an advantage of using SMT power devices on aluminum PCBs: The PCB is fixed to the heat sink with screws (with thermal compund in between). I use the fixing screws as a ground connexion to the heat sink, so the heat sink acts as a ground plane. .

Dear Francesc,

well, as I am aware of, the heatsink may be connected to a ground - but in most cases, it is not connected to power ground (of a half-bridge, f.e.) due to reasons of EMI. People try to minimise the capacitive coupling between system ground (PE) and the power ground which may float significantly.