GaN-on-Silicon Power Devices: How to Dislodge Silicon-Based Power MOSFETs
May 04, 2017
Gallium nitride (GaN) power transistors designed for efficient power conversion have been in production for seven years. New markets, such as light detection and ranging, envelope tracking, and wireless charging, have emerged due to the superior switching speed of GaN. These markets have enabled GaN products to achieve significant volumes, low production costs, and an enviable reliability reputation. All of this provides adequate incentive for the more conservative design engineers in applications such as dc–dc converters, ac–dc converters, and automotive to start their evaluation process. So what are the remaining barriers to the conversion of the US$12 billion silicon power metal–oxide–semiconductor field-effect transistor (MOSFET) market? In a word: confidence.
Design engineers, manufacturing engineers, purchasing managers, and senior management all need to be confident that GaN will provide benefits that more than offset the risk of adopting a new technology. Let’s look at three key risk factors: supply chain risk, cost risk, and reliability risk.
Supply Chain Risk
The list of companies selling GaN power devices is growing almost monthly. As of this writing, the list included the following companies:
- Efficient Power Conversion Corp.
- GaN Systems
- Dialog Semiconductor
- Navitas Semiconductor
- Texas Instruments
- ON Semiconductor
- VIS Technologies
- Cambridge Electronics, Inc
It should be noted that GaN Systems, Dialog Semiconductor, and Navitas Semiconductor all use a Taiwan Semiconductor Manufacturing Company, Limited (TSMC)-developed enhancement-mode process that is open to any fabless company wanting to enter this market. With all of these participants, and with the backing of major companies such as Texas Instruments, Panasonic, ON Semiconductor, and TSMC, the supply chain risk should be minimal.
Design engineers may delight in the performance of a system using GaN instead of a power MOSFET, but unless that system costs significantly less than its predecessor, there will be resistance from purchasing departments and upper management. Often component suppliers will make the argument that their costly product will save overall system costs. In my experience, this is a hard, but not impossible, argument to make. However, if the actual component cost is lower than the component being replaced, the argument changes. Instead of a design engineer having to convince upper management and purchasing of the benefits of the new component, it becomes upper management and purchasing asking the design engineer why he/ she has not yet converted. In May 2015, the Efficient Power Conversion Corporation’s (EPC’s) latest generation of product crossed over the price curve with MOSFETs of similar voltage and on-resistance. There were three main reasons for this change: 1) volumes had reached the point where economies of scale kicked in, 2) GaN transistor die sizes are three to five times smaller than MOSFETs, and 3) EPC’s eGaN transistors are in chip-scale format, which eliminates all the costs associated with packaging—costs that are 50% or more of the costs of a silicon MOSFET.
The area efficiency of current GaN transistors is more than 500 times lower than theoretically possible. It should therefore be expected that die sizes will continue to shrink, while performance increases and costs go down at each step.
There has been an enormous amount of reliability testing performed on GaN devices from several manufacturers, including EPC, GaN Systems, Panasonic, and Transphorm. It has been generally agreed upon that GaN devices can pass standard JEDEC testing originally designed for silicon-based power MOSFETs. The open question is whether these tests are adequate for GaN devices, given the radically different material.
Whereas we still have much to learn about all the failure mechanisms uniquely related to GaN technology, there is a growing body of evidence that suggests excellent reliability. Moreover, EPC’s eGaN transistors and integrated circuits with chipscale packaging do not have any of the reliability failure mechanisms attributable to traditional MOSFET packaging. Power MOSFET producers know that the majority of field failures are due to packaging-induced thermomechanical stresses or assembly problems. None of these mechanisms exist in chip-scale packages. As final evidence of basic reliability in real-world applications, EPC has tracked over 25 billion hours in customer applications over four years with only three device failures. This is a record unmatched by even the aging power MOSFET.
GaN-on-silicon has enabled many new applications in the seven years since volume production began. These new applications have helped develop a strong supply chain, favorable cost structure, and enviable reliability record. Forward-thinking companies are in various stages of adoption, while more conservative companies are waiting to see more compelling competitive reasons for conversion. In the long term, the performance and cost advantages of GaN-on-silicon will result in a majority of applications currently using MOSFETs converting to the smaller, faster, cheaper, and more reliable GaN technology.