Microwave Journal
www.microwavejournal.com/articles/38671-mmwave-power-amplifier-mmic-design-and-modeling-challenges

mmWave Power Amplifier MMIC Design and Modeling Challenges

August 12, 2022

Improved simulation accuracy is demonstrated using a hybrid electromagnetic (EM) device model, which includes EM wave interaction and coupling effects, versus a piecewise model for a power amplifier (PA) MMIC design.

The use of mmWave frequencies has surged in recent years due to the increased demand for wireless data. Both terrestrial and satellite communication systems continue to push higher into this frequency range to take advantage of the large available bandwidth. While both terrestrial and satellite communication systems will play a key role in our future wireless communication infrastructure, designing the MMICs to support these systems is often challenging. A key component of these wireless systems, and the most challenging to design, is the PA.

For high-power applications at mmWave frequencies, GaN has quickly become the semiconductor technology of choice due to its superior power, and more importantly, linear power performance. This technology also provides high gain and high efficiency while being extremely reliable. While there are many challenges to designing mmWave GaN PA MMICs, this article focuses on the growing challenge of supporting multi-system specifications and how to partition MMICs in simulation for accurate modeling.

MMIC DESIGN FLOW

Figure 1

Figure 1 MMIC design flow.

Figure 2

Figure 2 Measured IMD3 with 10 MHz tone spacing at 29 GHz, comparing the PA biased for QPSK (a) and 512-QAM (b) modulation and operating at 25°C.

The common design flow for developing a MMIC is shown in Figure 1. A design starts with collecting information from customers to understand their MMIC needs and system requirements. At this stage, the MMIC designer provides design input, and various semiconductor technologies are assessed to determine the best technology. The goal of this initial stage is to finalize the semiconductor technology and MMIC design specifications.

After these are finalized, circuit design software, which contains various types of simulators (e.g., harmonic balance, EM, thermal and others), is used. Passive and active models for the semiconductor technology are also obtained, most commonly from the semiconductor foundry in the form of a design kit. With these tools in hand, the MMIC designer performs analysis and optimization, in which circuit topologies are analyzed and optimized to achieve the design specifications.

Once the design is completed, it goes through various reviews before the start of fabrication. The cycle time to develop a MMIC typically ranges from six to nine months for design, fabrication and test, but this can vary greatly due to various factors such as circuit complexity and semiconductor foundry lead time. Given the time and cost to develop a MMIC, it is essential for designers to do everything possible to achieve first-pass design success.

SUPPORTING MULTI-SYSTEM SPECIFICATIONS

PA MMIC suppliers develop products to satisfy market demand for a specific frequency band and application. An example is the 27 to 31 GHz Ka-Band used for satellite communications. A portion of this band is also used for 5G mmWave applications.

Within the satellite communications market itself, the requirements for Ka-Band PA MMICs can differ greatly due to the variety of system architectures and designs. One of the main MMIC specifications that varies between systems is the maximum linear power specification, which relates to the linearity of the amplifier.

Linearity requirements are highly dependent on several system-level factors such as the modulation scheme, number of carriers and bandwidth. When different satellite systems have different specifications for PA linearity, it creates a unique and more complicated multi-system specification for the PA. The challenge is to create a MMIC that can support multiple systems, which is more cost effective and profitable for the MMIC supplier.

To illustrate this challenge, consider, for example, the linearity requirements for a QPSK modulation scheme versus a 512 QAM modulation scheme. For QPSK modulation, the maximum output power of a MMIC is commonly specified in terms of a spectral regrowth limit, most typically around -30 dBc. By rule-of-thumb, this would approximate to a two-tone third-order intermodulation distortion (IMD3) specification of -24 dBc. For 512 QAM modulation, the IMD3 specification will be around -45 dBc or possibly lower. Designing a single PA MMIC that provides optimum performance for both these cases is a challenge for the MMIC designer.

Designing PA MMICs to multi-system specifications requires flexible design approaches. One way is to design for multiple biasing schemes. By keeping individual stages of a PA MMIC biased independently and adjustable by the end user, performance can be tailored for different uses. Designing for different biasing schemes requires complex analysis and design work from the onset to determine the best transistor size and total periphery for each stage of the amplifier.

The purpose is to take advantage of the nonlinear effects, namely AM-to-AM and AM-to-PM, of individual stages where the nonlinear effects from one stage of a design can compensate for the nonlinear effects from another stage. This will change based on biasing; and the more gain stages used in a design, the greater the flexibility to accommodate multiple specifications. There are limitations, however, to the number of stages in a PA design, because too much gain can cause stability issues.

Table 1

An example of a MMIC designed to operate with different biasing schemes is Nxbeam’s NPA2003-DE. The NPA2003-DE is a 27 to 31 GHz 32 W GaN PA MMIC. Table 1shows the performance of this PA MMIC with the biasing scheme indicated in the figure. To showcase how this MMIC was designed for multi-system specifications, Figure 2 shows the IMD3 results for two different biasing schemes, specifically a QPSK modulation and a 512 QAM modulation.

Figure 2a is the biasing scheme for QPSK modulation. This biasing provides a nulling effect at around 42 dBm output power in which the nonlinear behavior of the second stage of the design compensates for the nonlinear effects from the third stage. By creating this nulling effect, higher output power can be achieved with a smaller output power back-off. From this curve, a power level of greater than 22 W can be achieved for a -24 dBc IMD3.

Figure 2b shows the IMD3 performance for the 512 QAM biasing scheme which is designed to provide a more traditional 3:1 IMD3-to-carrier ratio. It should be noted that the ratio for this MMIC is closer to 2.5:1 in the range of -45 dBc. The figure shows achievable IMD3 levels of -45 dBc or better required for this type of higher-order modulation. At an IMD3 level of -45 dBc, this MMIC provides 4 W of output power.

MMIC PARTITIONING FOR ACCURATE MODELING AND SIMULATION

Once the PA MMIC topology, device sizes and stage periphery ratios have been determined, the challenge moves to accurately modeling and simulating the circuit, which becomes more difficult at mmWave frequencies. Accurate modeling and simulation are critical to achieving first-pass design success. As previously mentioned, MMICs are designed using circuit design software that contains a variety of simulators. Passive components of a MMIC are typically modeled using EM simulators, while the active devices are independently modeled using a variety of different linear and nonlinear models. Some nonlinear model examples include Angelov, Materka and EEHEMT.

Since active device models are independently created, a piecewise design approach is usually taken to design a MMIC. Figure 3 shows an illustration of a piecewise design approach, in which different parts of a MMIC are modeled separately and connected within the circuit design software to simulate the combined circuit performance.

Figure 3

Figure 3 Piecewise MMIC design approach.

Figure 4

Figure 4 Section of the PA MMIC showing the connection between the passive network and active device.

During the initial phase of a design, lumped or distributed models may be used for the passive components but by the end of the design, EM simulation software is used as it has the capability to capture the true passive network behavior more accurately. With the speed and accuracy of today’s EM simulation software, the entire passive portions of a MMIC can be EM simulated. The final MMIC design will then consist of EM simulated networks of passive components connected directly to the active device models in this piecewise fashion.

To be successful with the piecewise design approach, it is important for the designer to understand how and where to partition a MMIC so that when connected in a piecewise fashion, the correct performance of the circuit is predicted. This partitioning is crucial at mmWave frequencies as passive and active components are moved closer together. Figure 4 illustrates this as it shows a portion of a circuit where an active device is connected to a passive network. The red dashed line represents a common plane to partition this circuit, however, understanding the effects of this partition can affect the outcome.



To understand circuit partitioning, it is important to look at how these individual piecewise circuit models are created, as well as the assumptions used. Figure 5 shows the passive network from Figure 4. It should be noted that much of the passive network has been removed to simplify this explanation. When simulating a passive structure such as this, it is important to understand what the excitation signal looks like on each port.

Figure 5

Figure 5 Passive network from Figure 4 used in the piecewise simulation.

Figure 6

Figure 6 Active device measurement test structure.

Figure 7

Figure 7 Hybrid EM active device model using grounded port (a) and internal distributed port (b) implementations.

In this case, standard port extensions are used on each port. The purpose of port extensions is to enable any higher-order modes developed from the port excitation method to decay before interacting with the passive structure being characterized. In this way, the resulting S-parameter file will be for a particular excitation mode, mainly the fundamental mode for each port transmission line. When developing a model using this approach, the MMIC designer must understand that the resulting S-parameter model is accurate only for this excitation mode.

Similarly, the development of active device models usually involves taking measurements of a device test structure, like the one shown in Figure 6. As can be seen from the test structure, relatively long transmission lines are used to feed the active device. In this way, the active device is also excited by fundamental mode excitation. Like the EM simulation, the active device model is accurate only for this mode of excitation.

The difficulty with mmWave circuits is that the distance between active devices and the passive structures is short, such that the mode excitation assumed when cascading individual models together may no longer be valid. This is the case shown in Figure 4. The proximity of the discontinuities from the shunt transmission lines close to the input of the active device will generate higher-order modes in that region of the circuit which includes the input of the active device. In this case, there is not enough distance between this discontinuity and the active device to develop a clean fundamental mode excitation. This will be referred to as EM wave interaction and this EM wave interaction must be accounted for in the simulation for accurate circuit prediction.

In addition to EM wave interaction, there are also EM coupling effects that must be considered when partitioning a circuit for piecewise simulation. If elements of the passive EM network are near the active device, this EM coupling may also need to be accounted for in the simulation. An example of this from Figure 4 is the close proximity of the top active device via to the nearest vertical transmission line of the passive network.

A well-documented method to account for this EM wave interaction and coupling effect has been to include more of the active device into the EM simulation.1-3 This is referred to as a hybrid EM device model. Hybrid EM device models have been around for over 20 years with many different styles and implementations. An example of two are shown in Figure 7. The goal of these models is to accurately capture the voltage and current waveform distributions on the active device manifolds or within the active portion of the device.

Figure 8

Figure 8 EM simulation of the passive network and hybrid EM device model.

In hybrid EM device models, internal ports are used within the active device that provide the terminals to connect to a core intrinsic device model, such as an Angelov model. Figure 7a shows an example where just the extrinsic device manifolds and source vias are included in the EM simulation. This EM simulation makes use of grounded ports to connect the active device model, while Figure 7b illustrates the use of internal distributed ports for connection to the intrinsic device model.1,2

The challenge with these hybrid EM device models lies in the implementation of the internal ports, which are not true to the real device and will cause their own inaccuracies. The goal however is to develop an internal port method that reduces these inaccuracies such that they have a negligible effect on the surrounding circuitry, or at least less of an effect relative to the EM wave interaction and coupling effects.

With the development of a hybrid EM device model, the EM simulation consists of both the passive structure and the passive portion of the active device model. An example of how the circuit structure presented in Figure 4 would be simulated using a hybrid EM device model is shown in Figure 8. Note that the other ports of the input passive network are not being shown for simplicity.

COMPARISON OF SIMULATION VERSUS MEASUREMENT

To demonstrate improved simulation accuracy using the hybrid EM device model and to show the effect of EM wave interaction and coupling effects, measurements of Nxbeam’s NPA2003-DE are compared with simulations using the hybrid EM device model approach versus the piecewise model design approach. It should be noted that this MMIC was designed using the hybrid EM device model approach. All simulations are done using Cadence’s Microwave Office Design Suite including AXIEM for the EM simulator. All active devices models are developed by Nxbeam.

Figure 9

Figure 9 Measured vs. simulated performance, using the hybrid EM device model for the MMIC PA: gain (a) input return loss from 27 to 31 GHz (b) and output return loss from 27 to 31 GHz (c).

Figure 10

Figure 10 Measured vs. simulated performance at 29 GHz, using the hybrid EM device model for the MMIC PA: output power (a), power gain (b) and PAE (c).

Figure 9 shows small-signal measurement of the NPA2003-DE versus simulation for just the hybrid EM device model approach. As can be seen from the gain in Figure 9a, the hybrid EM device model accurately predicts the gain and bandwidth of the MMIC with a slight exception in the high-end roll-off. The return loss comparisons in Figures 10b and c are included for completeness.

Achieving good agreement between measured and simulated return loss is much more difficult as the measurements are taken on a test module which contains additional substrates and transitions between the measurement reference planes and the MMIC. This can mask the true accuracy of the simulation methods. Calibration error can also play a role in comparing simulated to measured return loss at mmWave frequencies as measurement error typically increases with frequency.

Figure 11

Figure 11 Measured gain vs. simulations using the hybrid EM device model and piecewise device model for the MMIC PA.

Measured versus simulated performance using the hybrid EM device model for power is shown in Figure 10. The hybrid EM device model approach achieves close agreement in power, power gain and power-added efficiency.

To compare simulation accuracy between the hybrid EM device model approach and the piecewise model design approach, Figure 11 shows the small-signal gain result from Figure 9 compared with the piecewise simulation result. The gain from the piecewise approach is shifted approximately 1 GHz higher in frequency. In addition, the gain shows an upward slope across most of the band. If the piecewise design approach was relied upon for this design, the measured result would likely have been the mirror image of this, namely the gain shifted down in frequency from the desired band by approximately 1 GHz with the gain sloped downward in frequency.

CONCLUSION

The ever-increasing demand for wireless data will continue to push wireless communication systems to higher mmWave frequencies. Designing PA MMICs for these systems will continue to be challenging as many systems will have different MMIC specifications.

Designing MMICs for multi-system specifications will require more flexible design approaches. In addition, it has never been more critical to obtain first-pass design success due to the time, resources and cost. The hybrid EM device model approach, as well as other new models and methods, will be needed to support more accurate MMIC development in the future.

References

  1. A. Cidronali, G. Collodi, A. Santarelli, G. Vannini and G. Manes, “Small-Signal Distributed FET Modeling Through Electromagnetic Analysis of the Extrinsic Structure,” IEEE MTT-S International Microwave Symposium Digest (Cat. No.98CH36192), Vol. 1, June 1998, pp. 287–290.
  2. E. Larique, S. Mons, D. Baillargeat, S. Verdeyme, M. Aubourg, R. Quere, P. Guillon, C. Zanchi and J. Sombrin, “Linear and Nonlinear FET Modeling Applying an Electromagnetic and Electrical Hybrid Software,” IEEE Transactions on Microwave Theory and Techniques, Vol. 47, No. 6, June 1999, pp. 915–918.
  3. D. Resca, A. Santarelli, A. Raffo, R. Cignani, G. Vannini, F. Filicori and A. Cidronali, “A Distributed Approach for Millimetre-Wave Electron Device Modelling,” European Microwave Integrated Circuits Conference, September 2006.