GaN HEMTs offer power amplifier designers many improvements over LDMOS, GaAs, and Silicon Carbide (SiC) technologies. Desirable characteristics include high-voltage operation, high breakdown voltages, power densities up to 8W per mm, fT up to 25GHz, and low quiescent current. On the other hand, a GaN RF power device exhibits self-heating and a complex dependence of the nonlinearity of component parameters on signal level, thermal effects and ambient conditions. These factors often make it difficult to predict accurately the device large-signal performance.
To ensure device performance, test equipment is often used to measure device performance in desired applications, but this traditional approach has drawbacks: test hardware needs to be developed, and time-consuming load-pull measurements must be performed.
A large-signal model that closely matches the measured performance of the real device is preferred over physical testing for several reasons. It reduces development costs; it allows more in-depth ‘what-if’ analysis to determine if a device is the right fit before moving forward; it provides faster design cycles based on reduced characterization time and the ability to link layout optimization with end performance. The result is more first-pass design successes.
Inside Wolfspeed’s GaN HEMT Large Signal Model
Wolfspeed has developed extremely accurate 3-port large signal models for their GaN HEMT devices on SiC substrates, which feature high efficiency, high gain and relatively easy matching characteristics.
Figure 1 shows the large-signal model schematic and the intrinsic FET equivalent. The model is based on established equivalent-circuit methods. Data extraction is relatively simple; Wolfspeed uses a variety of test fixtures and test circuits, including load pull at the fundamental and harmonics. Large signal load pull and power drive-up is also verified at various frequencies and device sizes to ensure accurate large-signal scaling.
To be successful in scaling by a large ratio it is imperative that the unit cell model be extremely well correlated with measured data in all regions of operation. With an accurate and scalable large signal model in place it is then possible to design much large power transistors. The 3-port HEMT model has demonstrated success over multiple designs with scaling factors greater than 100 to 1. The non-linear model fits small-signal parameters over a range of bias voltages where measurements are performed using CW condition.
In addition to the three FET ports (gate, source, and drain), the models provide access to intrinsic drain current and drain voltage waveforms, as well as the junction temperature of the die. Intrinsic drain current and voltage waveforms are essential when designing complex PA architectures such as Class F, as they allow the designer to optimize device matching at fundamental and harmonic frequencies.
The model also includes built-in process sensitivity and non-linearity on individual elements as required. For example, the drain current source is the dominant contributor to device non-linearity. The gate current formulation includes breakdown and forward conduction, and all voltage variations of parasitic capacitances are derived from charge formulations.
An accurate package model is another critical element. A physically-derived modeling approach to the package parasitic interconnects has been developed which consists of a number of different tools including s-parameter.
Comparison of model data vs. measured data
Accuracy in modeling both small-signal and large signal behavior is critical.
Small-signal modeling is key for designers to predict gain, return loss, and stability of their power amplifier designs. Wolfspeed models are evaluated against measured data across different gate widths, finger numbers, and bias ranges to ensure model accuracy in all three critical domains: DC-IV, small signal, and large signal behavior.
Figure 2 compares the modeled data in red with the measured data in blue. The Smith chart on the left shows that the modeled data follow the measured data very closely in both magnitude and phase across different gate widths and bias values. The two results track very closely for a range of different current bias conditions.
Accurate modeling of the maximum gain Gmax is essential for designers to understand the maximum available gain for a given application and show how the device performs over frequency. The Gmax plot on the right is closely correlated with measured data.
It is essential that the model tracks IV behavior with VDS to properly characterize the large-signal performance of the device. Understanding the DC IV characteristics is a very important aspect of RF PA design.
Knee walkout and current collapse are common behaviors of GaN devices due to surface trapped charges during gate pulse switching. The modeled behavior is correlated against pulsed IV data as part of the model extraction process. Large-signal load pull and power drive ups are verified at various frequencies and device sizes to ensure accurate large-signal scaling. The load pull contours allow a designer to understand what impedance needs to be presented to the device to achieve desired power and efficiency.
The plots in figure 3 show that both modeled gain and PA efficiency follow measured data well beyond the 1-dB and 3-dB compression points. This is critical for GaN devices, since unlike equivalent LDMOS devices, a GaN HEMT tends to produce maximum rated power output beyond the 3dB compression point.
A useful video is available that discusses the use of the model to tune a PA application. Two examples are given.
When the power amplifier is tuned for maximum output power, the DC IV plot shows the load line passes through the maximum current of the device, 350mA.
A second application employs the model to tune for maximum efficiency. In this application the load line passes through the point where the device dissipates minimum power for acceptable output power. The result shows that although the output power has dropped by 1dB, the efficiency is improved by 15% compared to the maximum power case. The temperature port shows an increase of only 12 deg C as opposed to 60 deg C when it was tuned for maximum power.
In both cases the modeled data track very closely to the measured data.
To better understand the process for validating load pull data for the large-signal models, consult this application note.
Wolfspeed has developed large-signal RF models that demonstrate extremely accurate conformance to measured data. Wolfspeed’s Foundry business uses these models to ensure faster cycle times, higher reliability, and more first-pass design successes.
The benefits to designers include lower development costs, reduced PA design iterations, and an increased first-pass success rate.
The best news is that the models are free to qualified businesses. For more information, check out the four-part video series on Wolfspeed GaN RF large-signal models.