Dan Kinzer, chief technology officer at Fairchild Semiconductor, asks when to chose wide bandgap devices?
The promise of wide bandgap material is a tremendous gain in efficiency, size, and weight at similar or even reduced cost for power electronic systems.
Roughly speaking, wide bandgap devices can have 10 times better conduction and switching properties. That gain is critical for pervasive adoption of wind and solar energy, hybrid electric and electric vehicles and power grids with distributed energy generation and storage.
What needs to happen to further the adoption of wide bandgap devices? What is the future for silicon power devices?
First, consider silicon carbide (SiC). Research on SiC power devices has been ongoing for more than 20 years. The first commercial power devices emerged 10 years ago as 600V Schottky diodes, which then progressed to junction barrier and merged PiN Schottky diodes.
The great thing about silicon carbide devices is the robustness that can be achieved due to the material. The thermal conductivity is more than x3 better than silicon, and the homogeneous substrate and epitaxy layers allow for vertical power devices that can spread heat generation uniformly across the die, with high current surge and high transient voltage and power capability.
Just now, SiC transistor switches are making their first commercial appearance. Early mosfet offerings have relatively high on-state resistance, and still have limitations due to surface channel mobility and gate dielectric stresses.
New mosfets are in development and partially address these limitations. Bipolar and JFET devices avoid these issues, and as a result have lower resistance that can approach theoretical limits.
Bipolar devices are normally off, which is a must for most applications. New BJTs have no storage time, 20ns switching, high current gain above 100, no gain rolloff at high current, negative gain and Vce(sat) temperature coefficient for stability, and no secondary breakdown. JFETs can be fabricated to pinch off at zero gate bias, but this increases the on-state voltage, and limits the gate voltage swing to less than 3V.
Normally on JFETs have conduction losses almost equal to bipolars, but usually require a cascode device in series, or at least a negative drive if the normally on operation is tolerable.
What about gallium nitride (GaN)?
Lower voltage parts ranging from 30V to 200V have already been introduced. Below 100V, the devices actually have higher specific on-resistance than the silicon alternatives.
A small advantage in gate charge is the only edge these devices have, while driving the device without gate rupture can be challenging.
Commercial 600V devices are about to emerge. Some of these are built on silicon carbide substrates, but for cost reasons most companies are focusing on GaN epitaxy on silicon substrate approaches.
This is an incredibly tough process challenge due to the large crystal mismatch; MOCVD processes are in development that can deliver several micron thick layers, enough to handle the 600V requirement, without excessive warping or cracking of the layers.
The inevitable dislocations that arise due to the mismatch, usually in the range of 109/cm2, need to be suppressed to avoid leakage to the conductive silicon substrate.
It is important to incorporate special impurities in the films to control the leakage as well as bulk charge trapping. Surface and bulk charge trapping can lead to on-state voltage increases and blocking voltage instability. Fortunately, a lot of progress has been recently reported on addressing these instabilities.
In theory, a vertical device in GaN could have better conductivity than SiC. This is often shown on specific Rdson vs rated BV graphs. The problem is the lack of a homogeneous GaN substrate at a reasonable cost and diameter.
Consequently, almost all efforts are on lateral high electron mobility transistors (HEMTs), which do not follow the model for vertical devices. The performance of these devices depends on reducing feature sizes, 2DEG contact resistance, and the drain drift length. This means high surface electric fields are unavoidable to achieve low resistance and these devices are not able to withstand significant avalanche current.
The devices must be overdesigned to ensure no voltage transient ever reaches the actual device breakdown voltage. HEMTs are normally on devices with a leaky Schottky gate, so for high voltage an insulated gate structure and an innovation in normally off device design is essential.
The key to success in silicon carbide is to accelerate the cost and material defectivity learnings, expand substrate and epitaxial capacity, and to transition to 150mm diameter to access widely available wafer fabrication.
Expect to see this happen in the next 2-3 years commercially for 600V to 1700V devices, with much higher voltages possible as well. The key to success in gallium nitride is improved high volume and lower cost MOCVD processes on silicon in the 150mm to 200mm range, with device and material designs that can withstand the high operating voltage and surface electric field stresses.
This has begun to happen in development for 100V to 600V devices, so expect to see these devices ramp in 2-3 years as well.
Will silicon devices withstand the onslaught? Absolutely! With decades of proven reliability and field use, and a highly mature and cost effective manufacturing infrastructure, IGBTs, silicon mosfets and rectifiers will serve the 600V to 1200V market well for years to come.
As system designers learn to use the high frequency capability of the wide bandgap devices, the system performance, size, and cost advantages will emerge and drive a gradual shift in the industry.