FAQ
GaN Thermal Analysis for High-Performance Systems
1. What are the challenges and considerations involved in GaN thermal analysis for high-performance systems?
GaN products are capable of producing significantly higher power than GaAs or Si, with very high power densities that
require a precise library of material properties and a solid understanding of transistor thermal behavior. The impact of
boundary condition assumptions is also much more significant – particularly legacy thermal boundary condition
assumptions for customer environments, which are often overly optimistic. An example is the suggestion that a packaged
product will be attached to a cooler heatsink than will actually be achieved.
TriQuint has developed a multi-prong approach to GaN thermal analysis that allows us to fully appreciate the thermal
implications of GaN technology. The approach includes:
Thermal modeling and empirical measurements, including micro-Raman measurements
Thermal analysis including FEA
Infrared – with an understanding of its limitations in this context
RF testing
Die-attach performance, including accounting for contact resistance when comparing epoxy to solder
Package and package attach options
2. What are the uses and limitations of infrared in modeling heat generation?
Infrared microscopes are widely used for determining fault location by searching for hot spots in semiconductor devices.
However, the application of IR for thermal characterization of FETs is limited due to spatial resolution incompatibility. IR
microscopes cannot resolve a spot size as small as the active area of a GaN transistor, meaning that IR measurements
average-in colder, non-active areas with the active area of interest. Additionally, peak temperature conditions of GaN
transistors occur below the surface of the substrate and cannot be imaged by infrared thermography, which is limited to
measuring surface temperatures. Emissivity must also be considered, as it is variable and difficult to control, and can lead
to erroneous data.
As an example, when an IR measurement is taken of an area that is only 0.25µm wide, the resultant temperature reading
o
can be 20-30 C cooler than the actual peak temperature of the active area.
3. What is the best approach for accurately estimating die-attach thermal performance?
One of the challenges of thermal modeling is finding an accurate estimate for die-attach thermal performance, which is a
very important factor in device operating temperatures.
Vendors of die-attach materials often only list the bulk thermal conductivity (k) of their product, but this is only one
component of the overall die-attach thermal impedance. Bondline thickness, interfacial resistances, voiding, and filler
characteristics all contribute to the overall thermal resistance, and these are to a large extent dependent on dispensing
and curing processes. Also, die-attach integrity and performance are impacted by material properties and surface
characteristics of the two items being bonded. Experimentation is usually required to know with reasonable certainty how
a die-attach solution will perform.
Illustration of Die-Attach Thermal Resistance / Impedance
The above graphic illustrates the relationship between bondline thickness,
bulk thermal conductivity and total die-attach thermal resistance.
Illustration of the Importance of Accurate Die-Attach Performance Data
o
For this particular die / package combination, TCH is underestimated by 40 C when interfacial
resistances are not taken into account.
4. Regarding die-attach methods, do epoxy and solder compare?
When comparing conductive epoxy to solder, contact resistance is often neglected, and it is assumed that the thermal
conductivity of an epoxy joint is equal to what is represented on the epoxy datasheet. Solder is significantly less impacted
by contact resistance and consistently provides better results in application for high power die.
5. Have packaging options kept pace with GaN development?
As GaN technology has developed, and its power potential realized, existing semiconductor packaging technology has
proven incapable of effectively supporting high power GaN products. In response, TriQuint has developed improved
packaging options that provide a complete product solution for the ever-widening array of military and commercial GaN
applications. Included in TriQuint's new GaN package options for high-performance systems are GaN on Copper and GaN
in Plastic.
GaN on Copper refers to a copper flange-type package with leads (commonly called a module). This packaging option
simultaneously increases system level reliability and allows for solder-attach to a high CTE heat sink.
GaN in Plastic refers to an air cavity plastic or over-molded plastic. Advantages include pulsed operation performance,
very small size and low cost. Plastic packages are CTE-matched to PCBs, so system level reliability is excellent.
6. Does the use of plastic instead of metal ceramic in packaging affect heat extraction?
Heat removal must be considered throughout the entire board. This starts with die attach, because with plastic packaging,
there is typically a copper alloy material used for the lead-frame. This offers advantages in heat removal, as well as CTE
match to the board.
The package has increased thermal performance because of the copper lead-frame, so attention should be paid to the
PCB to ensure that it is a thermally enhanced PCB, includes thermal vias, or has a solid metal/copper slug in order to
achieve the full potential of the package.
The bottom line is that plastic packaging provides better heat removal than a ceramic package.
7. What does the future hold for packaging GaN devices?
The industry is moving to copper and plastic. Whether it’s a plastic package with some sort of copper alloy for the leadframe or using copper in flanged packages – those are the two next-in-line materials that promise to increase heat
transfer and decrease cost.
Further out is diamond and diamond-composite materials. These have higher costs and are still in development phases,
but they do have a future and offer significant thermal benefits for the right applications.
8. What are the main focus areas for TriQuint when it comes to attaching a die to a package?
For the plastic packages or small devices, epoxies are maturing to handle higher heat flow while still being able to absorb
the mechanical stresses that result from placing a silicon carbide die into a classic copper alloy surface mount package.
So currently the main focus is on different types of epoxies or dispensed die attaches.
For larger, higher power devices, TriQuint is using AuSn die attach. Here, it is critical to use highly controlled processes
and optimized tooling per device types for best performance.
9. What are some important factors when modeling and measuring heat generation and removal?
When modeling and measuring heat generation and removal, it is critical to select proper boundary conditions and to
understand the impact of those assumptions. Commonly, unrealistic or improper assumptions are made on temperature
and heat removal boundary conditions. The unrealistic predictions and measurements that result often lead to product
designs that appear to work well on a datasheet, but will fail in application.
10. How does emissivity factor into thermal analysis?
Emissivity (ε) on the surface of a die varies significantly. The common solution for this is to paint the die matte black to
achieve an assumed ε = 1. The challenge this creates is that paint imparts a dielectric load on the die, which can be
difficult to predict and is typically inconsistent. This changes RF performance and as a result, accuracy and
repeatability suffer.
11. Why do maximum channel and junction temperatures seem to vary so much between GaN suppliers?
From a manufacturing standpoint, different wafer fabs simply have different processes, as well as different material layers,
and different materials themselves, all of which can create more or less sensitivity to temperature.
More important is the fact that different suppliers specify different failure criteria. TriQuint is more stringent than most,
allowing only about half as much degradation as other suppliers. Also, measurement and prediction of channel
temperatures creates a significant opportunity to introduce error. Again, though, TriQuint sets – and meets – extremely
high standards for accurate measurement.
12. Is there a need for improved thermal management materials? If so, what thermal conductivity is needed?
There certainly is a need for improved thermal management materials. The thermal conductivity needed requires an
examination of the trade-offs between cost and performance, i.e., the ultimate application or system, the cost targets,
power level requirements, and the needed level of sophistication of heat removal. If the absolute best thermal
performance is required, regardless of cost, then diamond would be the preferred material. If the requirement is thermal
improvement over a current standard while maintaining a similar cost, then copper could be the choice for a significant
improvement in thermal performance with very minimal cost change.
13. Would a 2mil wafer improve thermal performance over 4mil?
No, it would not. TriQuint has simulated this, and it actually hurts thermal performance. The reason is that silicon carbide
is a very good thermal conductor, and the extra thickness of SiC allows heat to spread laterally before it hits the
resistance created by the die attach, so there is less heat flux at the die attach. This means that the density of heat flow is
decreased by the time it gets to die attach, and the result is a distributed heat load instead of a focused one. This reduces
temperature rise through the die attach joint, and it also reduces the sensitivity to voiding at die attach. The positive
effects of lateral heat spreading in the SiC also apply to materials below the die attach. So from a thermal perspective, the
thicker SiC is actually a good thing.