Thursday, March 28, 2024
Login

RF/Microwave EDA Software Design Flow Considerations for PA MMIC Design

Parent Category: 2021 HFE

By Cadence AWR

The evolution of integrated circuit (IC) technology demands that designers in this field adapt to ever-changing manufacturing techniques driven by performance, cost, benefit, and risk considerations. Today’s PA designers working in solid-state technologies must navigate a plethora of available processes, including GaAs, gallium nitride (GaN) and silicon carbide (SiC) PHEMT, radio-frequency complementary metal-oxide semiconductor (RF CMOS), and GaAs or silicon germanium (SiGe) heterojunction bipolar transistor (HBT), to name just a few.

Similarly, different design challenges demand different amplifier classes and/or topologies like Class AB, Darlington multi-transistor configuration, switch-mode PAs, and digital predistortion. Moving from one technology to another implies that certain skills and knowledge are transportable and transferable. The most basic of these skills is the effective use of electronic design automation (EDA) tools for designing the MMIC itself. More specifically, it is a strategy, design flow, or guidelines for how to begin with requirements and a process design kit (PDK) and arrive at a point where the more complicated requirements can be tackled.

In this white paper, a gallium arsenide (GaAs) pseudomorphic high-electron mobility transistor (pHEMT) power amplifier (PA) design approach is examined from a systems perspective. It highlights the design flow and its essential features for most PA design projects by illustrating a simple Class A GaAs pHEMT monolithic microwave IC (MMIC) PA design using Cadence® AWR® Microwave Office® circuit design software. The paper begins with the concepts of design closure and then describes parametric design as key concepts to understanding each step of the PA design process.

The Design Flow in General

Design flows can sometimes appear to be a chaotic affair, but there is a logic and order to walking a design through a process from concept to completion. Design flows can be viewed in several different ways, and perhaps this is what causes some of the confusion. If viewed as a series of steps that are repeated or iterated until the simulated performance converges on the desired requirements, it’s easy to miss much of the underlying structure of the design that makes it a repeatable, reliable methodology. Top-down design flows are highly desirable as they provide predictability. By relating design parameters to overall performance, engineering teams can drill down into the components with cause-and-effect relationships clearly defined.

Bottom-up design, on the other hand, assures to some degree that each individual piece of the design contributes its necessary functionality as imagined by the engineering team. By showing that each component does this from a micro-to- macro, netlist-to-behavioral, layout-to-schematic, etc. perspective, the engineering team demonstrates that its design converges to the requirements; they’ve achieved “design closure.” These, then, are two of the more substantial hallmarks of the engineering process: parametric design and design closure. A very succinct treatment of design flow is described that has been elaborated on many times elsewhere for microwave [1-4,6] and analog mixed-signal [5] designs.

The Design Flow as Applied to a GaAs MMIC PA

For the typical GaAs PA design, the design flow plays itself out in Figure 1 Required performance leads to a sequentially more detailed design step (dotted outline box), where the design team strives to define more and more of the PA’s behavior as more complex phenomena are explored. The detailed understanding of the circuit gleaned in each substep of the design ultimately secures a complete picture of the PA’s complex performance—in essence a “walk before you run” approach.

2021 05 design 01

Figure 1 • Generalized incremental design flow with iterated analysis.

 

The first iteration may only be selection of the bias point, but even here there can be a substantial amount of complexity, such as load-pull or thermal considerations. The second pass through the design flow will focus on linear performance and stability in terms of input and output network design. Again, this should not be trivialized, as a large linear array of devices must be fed with manifolds that not only provide good matching to the source and load impedances, but also feed all the individual field effect transistors (FETs) in the array in-phase. Linear design is followed by nonlinear performance, and this is where the design really gets down to business; saturation and efficiency are examined in detail while still assuring linear gain and stability.

Tradeoffs must be made and then fine-tuned as the layout is produced and then extracted back into the simulation for the fourth and final design substep. The analysis follows the design to assure that assumptions are justified, and second order effects are understood. Issues commonly dealt with here include a full electromagnetic (EM) analysis to assure layout standards and minimized couplings, and an iterative electrothermal analysis to guarantee that the channel temperature has been adequately modeled. The final step, verification, is used to prepare the design for manufacturing with design rule checks (DRC) and provide one final opportunity to ensure that the design is compliant with performance, manufacturing, test, and packaging requirements before tapeout.

PA Design Example

Step 1: Requirements

Turning to the GaAs PHEMT PA design flow, this example examines the role of various design considerations and shows how parametric design and design closure manifest themselves. The assumed specifications are for a Class A amplifier powered by a VDC source that is summarized by maximized power-added efficiency (PAE) for a given linear antenna output power, Pant, slightly backed off from the PA’s P1dB with moderate (single stage) gain (G), and some antenna mismatch.

Step 2: Bias Selection

Substep 2.1: Power Dissipation

To realize this design, users can immediately dive into the bias selection with some quick calculations. In this first design substep, as in all steps, it is necessary to clearly articulate what design requirement is to be achieved, identify the design parameters that dominate in determining how the design meets this requirement, and then show that the selected values for these parameters satisfy the requirement.

Without selecting a proper bias point and associated FET periphery, there is no gain (G), the amplifier may not support enough output power (Pant), and it may be far from linear (P1dB). In short, if the PAE is used as a design requirement and the pHEMT DC Ids and Vds are viewed as design parameters, then this first step can be viewed in terms of parametric design.

For the actual PA output power, the other components in the design may need to be considered. Sometimes, PA design requirements are not given solely with regard to the PA itself, but instead are specified with regard to the system, or both the system and the PA. Figure 2 shows a reasonable system diagram from the PA to the antenna and highlights several additional components that may need to be considered in translating the system requirements into those for the PA.

The inclusion of switches and a consideration of the antenna impedance create a relationship between the power at the antenna or the system power output and that which is required of the PA.

Pout= Pant + Lossswitches + Lossmatch + margin (1)

Lossswitches is assumed as the loss through a transmit-receiver (TR) switch and/or diversity switches (Figure 2), Lossmatch as the mismatch at the antenna, and margin as a combined design and back-off margin. While in the ideal world of computer-aided design (CAD) simulations, the PA will deliver all the power it simulates; in reality, when built, users would expect there to be device-to-device or lot-to-lot variation in Pout for which they would wish to account in the margin.

2021 05 design 02

Figure 2 • System diagram showing components contributing to specification of PA based on performance at the antenna

 

From (1), the power dissipation (PDC) within the PA can be determined from the PAE:

PDC = exp(Pout/10) * 1/PAE (2)

With a reactive load, the drain-source voltage and VDC are roughly identical—in other words, in the absence of a drain (load) resistor, the DC drain bias is dropped entirely across the transistor drain source.

IQ = PDC/VDC = IQ(Vgs, Temp) (3)

From IQ, the quiescent drain-source current, the IV curves for the device, and the corresponding gate-source voltage (Vgs) at which the PHEMT should be biased as a function of temperature can be consulted. The FET periphery is sized such that selection of Ids as a function of Vds is approximately halfway between VDC and the “knee” of the IV characteristics at the channel temperature. The choice of “halfway” is determined by the desire to have a Class A amplifier and will change depending on whether an AB, B, etc. topology is chosen. This gives the essential aspect of the first substep in the design.

This entire substep can be accomplished quickly and efficiently with AWR Microwave Office software. A DC IV sweep can be set up using either one of the two IVCURVE elements to simulate a nested DC sweep (voltage over current for bipolars or voltage over voltage for FETs). Most well-supported MMIC processes include FET models with DC bias over temperature, so the IV curves can be further explored through tuning/sweep. A graph of classic FET IV curves cast as IDS versus temperature (shown in Figure 3) can also be helpful to see the current gradient and whether it is substantial.

2021 05 design 03

Figure 3 • IV characteristics at ambient (blue, 25°C ) and elevated (red, 100°C ) operating temperatures.

 

Substep 2.2: Thermal Dissipation

Before actually moving on to the small-signal design, it is not a bad idea to pause and reflect on the thermal implications, especially since there is a temperature dependence in Equation (3). FET devices like GaAs PHEMTs are majority carrier devices with the control terminal dominated by a voltage determining the current at the output. Thermal sensitivity to this physical process comes from increases in the majority concentration, which are more than offset by additional scattering mechanisms that in aggregate manifest themselves as a decrease in output current with increasing operating temperature. This is a negative feedback process. The balance between these two effects is clearly illustrated in Figure 3. At lower currents, the warmer device has more carriers in the channel and the resulting current is higher than for the room temperature device. As self-heating becomes a dominant factor, the elevated temperature increases the scattering, making it more difficult for carriers to traverse from source to drain, thereby reducing the current flow.

In practice, the pHEMT current changes relatively slightly [6] and, aside from very real reliability considerations, temperature is less of a factor in the pHEMT design than in the electrical design of a PA using a bipolar junction transistor (BJT), where thermal runaway is a very real concern. From a design flow perspective, since the bias is sensitive to temperature and the temperature in the FET channel will tend to “pull” the bias one way or the other as compared to room and/or baseplate, it is prudent but not necessarily essential to consider temperature as part of this initial bias design step.

Substep 2.3: Load Pull

Another consideration in bias selection of PAs is load pull, or the shifting of the effective output impedance of the FET in its nonlinear operation. This implies that as the input signal power is varied, the FET will operate in a linear fashion at lower powers, but then shift as the power is altered. In reality, the load impedance is altered while measuring a particular performance parameter so that the impedance presented to the FET can be advantageously chosen. Alternatively, given that the FET operates with some degree of nonlinearity, how is this nonlinearity altered by the load impedance? This is clearly a parametric design flow issue as there is specified performance criteria tied to a design parameter—the load as seen by the FET.

Thus, when considering load pull, it is not enough to say that a bias was selected based on load-pull considerations—what nonlinear output characteristics are being “pulled” by what load impedance must also be stated. For this reason, load-pull data is often presented as circles on a Smith chart. PAE or saturated output power are typical values, but second- or third- harmonic cancellation can also be important. In more detailed approaches to PA design, such as waveform engineering, the entire FET model is essentially the load-pull data for the FET using a close facsimile of the desired input signal. For these reasons, this step might be considered as part of Step 3, linear design, since the performance criteria being monitored is something other than the IV curves.

The AWR Microwave Office load-pull utility is a great script to invoke for doing this in the context of the design. Figure 4 shows load-pull contours for the device used in Figure 3—contours of the PAE are plotted against two different bias conditions. This allows the designer to check the efficiencies available given the DC dissipated power (i.e., different bias conditions) and different (load) impedances presented to the transistor (i.e., the load-pull contours themselves). Perhaps more important from the design perspective is the use of the simulated load-pull capability simultaneously with the conjugate small-signal output match of the transistor versus gate bias (Figure 5). Since maximal power transfer requires the conjugate output match of the transistor, this graph is of key importance for the design flow as the cross section of maximal power transfer and PAE make for a very good PA. One note: Since this PA has been specified to operate in a mode “backed off” from P1dB, the small-signal S22 can be used in lieu of a measurement that gives the large-signal equivalent.

2021 05 design 04

Figure 4 • Simulated PAE load-pull contours using the AWR Microwave Office load-pull script to compare two different gate bias conditions with constant drain bias.

 

Again, as with the thermal “mini-flow,” if the time is taken to set up a load-pull analysis of just the FET stage, this can be used as a design closure condition for the remaining steps. It can also be expanded later on to look at nonlinear performance as well as confirm FET stage performance when the impedance of the output matching network is better understood.

Step 3: Linear Design

Without oversimplifying, the next step is to get the bias and linear performance in order by adjusting the parameters defining the input and output networks presented to the pHEMT. In other words, the parametric design aspect of the flow enables the user to control the linear performance by adjusting the input and output impedances seen by the FET. In advanced flows, this may mean designing the input and/or output network at not only the fundamental frequency of the PA, but at the harmonics as well. Design closure, the other flow criteria, is reached by maintaining the performance in substep 2.1 (essentially a DC bias that stays close to achieving P  and Pout) while obtaining the gain and match as required.

2021 05 design 05

Figure 5 • Simulated PAE load-pull contours versus conjugate S22 match.

 

Typically, the pHEMT periphery obtained from getting the correct bias and meeting the output power constraints gives input and output impedances near 50Ω, but perhaps not close enough, so impedance matching to some degree may still be required. With a large FET, the input gate-source capacitance can be fairly large (Figure 6), so as the frequency of operation goes up, the input impedance will start to vary more substantially with frequency and the input match becomes more challenging.

The input match should be implemented with an eye to stability and temperature is a minimal consideration in this substep. Specifically, if the DC bias network for the gate of the PA FET is properly designed with chokes and bypass capacitors, there is the possibility for creating low-frequency resonators, which could lead to oscillations. So, in addition to having the gain (G) as the design requirement for this step, linear stability indicators, such as K and B1, should also be included. In more advanced design flows stability would also be included as a nonlinear design goal [8].

As important as the input match is to “playing nice” with the earlier components in the transmitter chain, the bread and butter of the PA is the output side. First and foremost, without the output network being properly designed to give gain, this won’t be an amplifier, let alone a PA. To get the most voltage swing, and hence highest power out, an inductor for the load is used so that there is minimal resistive loss limiting the Vdc voltage available from that which can be seen at the FET drain. In other words,

Vd=Vdc-min(Vds(t))-IdsRe(ZL) (4)

or, the voltage available at the drain will be the DC source voltage less the minimum Vds necessary to keep the FET from going over the knee and the voltage from dropping due to any real impedance component in the load. At first blush, the answer would be a big MMIC inductor; however, big MMIC inductors normally come with larger resistance, so there is a tradeoff to be made. This tradeoff must be considered whether the PA is fully monolithic (with load inductor on-chip) or provided externally. The external, off-chip inductor can be quite attractive because of the higher Q and lower loss available, but the extra parasitic capacitances and inductances in getting to the off-chip device introduces further stability concerns.

As alluded to earlier in the discussion on load pull, a good output match is also essential for staying within the margin specified in (1) and should be identified through load-pull simulations or measurements [7] to identify the needed conjugate match for maximum power transfer:

Zout = Z*d (5)

where Z*d is the complex conjugate of the impedance looking into the FET/load circuit and Zout is the impedance looking into

the output matching network from the FET/load circuit and terminated at its output by the desired load (typically 50Ω).

This point in particular cannot be overemphasized as Pout becomes more substantial. The role of the PA is to get power to the antenna and the easiest way to miss this goal is to have power being stored or dissipated between the PA FET and the pin or connector representing the PA. The design of a proper matching network that transforms the “not exactly” 50Ω at the FET drain to the “as close to” 50Ω as possible at the pin/connector can be the difference between marginal and extraordinary efficiency.

Design closure for this step should confirm that the DC bias still provides the nominal DC power consumption for the expected PAE and that the linear gain (G) and any input and output matching criteria (in terms of VSWR or S11 or S22, respectively) has been achieved.

Implementing this in AWR software is fairly simple and is no different from designing small-signal amplifiers or passive circuits. Create the schematic, implement the desired measurements on several graphs, and go. One hint to speed things along for later steps: The circuit can be set up in hierarchy with testbenches at the top level, especially for the nonlinear simulations, but be sure to keep the initial schematics simple by using ideal components and reserve the MMIC PDK spiral inductor models, for example, until the topology is more certain. This tends to make the key design parameters easier to identify early on because there’s no confusion of results by parasitic effects.

2021 05 design 06

Figure 6 • The PHEMT presents input and output impedances not matched to 50Ω and quite capacitive.

 

Working from Figure 6, it can be seen that the S22 for the transistor is nearly on the real part of the impedance circle corresponding to 50Ω. (Figure 5 also shows this value as the conjugate match in the upper half of the Smith chart). This makes a conjugate match appear to merely require adding an equivalent series inductance, but this gets progressively more difficult at higher frequencies (since the length of the interconnects turns the inductor into a transmission line) and the power output goes up (since the lines will need to be thicker to handle the current in accordance with the design rules). Normally this is more complicated, and some stepped impedance transformer is needed to get the real part of the impedance to be matched as well as “matching out”—or conjugate matching—any of the imaginary part. For this design, an equivalent inductance of a few nanohenries (Figure 7) suffices, but how to implement this is a task for the layout, if it is to go on-chip, or for the packaging, if it’s just not feasible given the size and current handling constraints already mentioned.

2021 05 design 07

Figure 7 • Matched output impedance using a lumped inductor to give the equivalent conjugate impedance to the transistor S22.

 

Referring back to Figure 5, note that the match for maximum power transfer does not correspond to the match for optimal PAE. A tradeoff will need to be made when the nonlinear aspects of the design are considered, but at this point, we have achieved design closure in the small signal by completing the small-signal design criteria, namely the match that we would like.

Step 4: Nonlinear Design

This third trip through the design parameter/closure loop focuses on the nonlinear performance, PAE, and P1dB by fine-tuning the bias and match. But just as in the linear design step, in order to achieve design closure, what’s already been gained should not be disturbed, so the nonlinear performance must be optimized without sacrificing gain, match, and, perhaps most importantly in this step, stability. It will be very tempting to boost the PAE in ways that undermine the stability of the design—after all, what better way to get more power out with the same DC power than to create an oscillator!

From the perspective of the design flow’s parametric design requirement, it may seem like the design parameters controlling the performance of this step are the same as the previous step, except the nonlinear simulator is used to look at PAE and P1dB (or some other measure of nonlinearity). But this is also an ideal step to expand the consideration of what actually comprises the input and output matching networks to include bias lines, grounding, and bondwires or bumps that are off- and on-chip. Typically, bounds on parasitic source inductance are monitored to give guidance to layout (bound-pad number and placement) and packaging (bondwire count and length) in regard to not only degradation of the nonlinear performance criteria, but also to assuring that the requirements associated with the previous substeps are still being met.

The focus of this design step is to try to push out the compression of the linear output power as the input power is ramped up, as well as to boost the PAE. Strategies for doing this—thereby achieving this steps’ design closure—will focus on minimizing parasitics and adjusting the bias conditions slightly. It can be tempting to change the FET periphery, but this can be dangerous, especially if load-pull has been involved, since the linear part of the design has presumably been optimized based on a detailed understanding of the FET input and output impedances. If the IQ from (3) can be backed off to boost PAE without jeopardizing linearity, then some thermal margin of sorts is created with the lower current.

One note should be emphasized in regard to FET modeling. A clear and detailed understanding of the FETs’ nonlinear behavior—and to what degree the models being used capture those nonlinearities—is essential. For example, if the intent is to minimize third-order harmonic generation by clever impedance matching as a way to extend P1dB, then the model being used should not only be accurate in its ability to generate the third harmonic via derivatives of gm (current-based nonlinearity) or Cgs/Cgd (capacitance-based nonlinearity), but also should do so with the load impedance being something other than 50Ω. Such demands on a model are not trivial, and, conversely, trying to simulate and “design out” such behavior without validating that the model is so capable is a waste of time at best and foolhardy at worst.

If the testbench style of project organization (Figure 8) has been done, the AWR Microwave Office nonlinear simulations can be reused with linear analysis simply by changing the measurement being performed on the testbench—ports in AWR software (even nonlinear source ports) act as S-parameter terminations so dual use can be obtained from a graph. In AWR Microwave Office software, the port elements double as both linear terminations and subcircuit/hierarchy elements. Reuse of the underlying schematic across all analyses is important if the testbench’s underlying schematic(s) begin(s) to include off-chip or bias-related parasitics. Furthermore, if the PA is Class C or better, the designer can start using transient analysis with this same simulation set up at the testbench schematic level.

2021 05 design 08

Figure 8. Testbench style of project development where the subcircuit is shared among (left) linear and (right) nonlinear analyses to ensure consistent capture of parametric design and design closure criteria.

 

For this particular design example, the nonlinear measurements of importance are the PAE as well as the gain compression (Figure 9). The PAE is the parameter that needs to be optimized, but the original constraint that was introduced for this design spoke of having the PA “backed off” from P1dB. This means that the PA’s actual operating point for some given output power must correspond to an output power that is lower by some margin from the point where the amplifier gain is seen to compress.

2021 05 design 09

Figure 9 • PAE (%) and output power (dBm) for the transistor optimally matched for maximum PAE versus the transistor conjugately matched for maximum power transfer.

 

If this particular FET size and bias are used with the conjugate match for maximum power transfer, then the PA compresses at lower output powers and it does not deliver the optimal PAE. This can be understood from the load-pull contours in Figure 5, since the intersection of the conjugate S22 match with the load-pull results for this bias point show that the PA will not achieve the PAE maximum. The load-pull contours at this bias clearly show that both cannot be achieved simultaneously.

In practice it is more often the case that the PA design will demand the conjugate matching of the transistor at the expense of PAE. In this case, the load-pull would have been relative to maximum power transfer and not to PAE. This would have yielded an optimal match that would not have corresponded to the small-signal S22 but would have given an impedance that “pulled” away from the small-signal S22.

Step 5: Extracted Layout

Having completed the electrical design, the final design step is to actually lay out the PA. The parametric design requirement is somewhat lost at this point if the interconnects are not captured on the schematic, so, to the greatest degree possible, microstrip or coplanar waveguide elements should be placed on the schematic so that lengths and widths can be tied to maintaining the overall chip performance criteria. More than a few MMIC designs, PAs included, have never made it past this stage of product development simply because this parametric design requirement was lost at this stage in the design process in an endless series of “move a line, run the EM solver, simulate the circuit....and repeat.” In the face of dozens or a hundred interconnects and an extracted layout that does not achieve design closure, it is essential that the design team ascertain which interconnects control critical MMIC performance at the earliest possible time. If the MMIC PDK supports bond pads, then these should be included in the parametric design stage as well.

Design closure is achieved when the nominal simulations, including all these effects, confirm that the overall chip performance criteria have been met. Small (less than chip scale) EM analysis can be done locally to confirm that the input and output matching networks achieve their desired performance, such as that defined by equation (5).

The flow in AWR Microwave Office software is a bonus when it comes to this stage of the PA design flow. Using EXTRACT technology along the lines of the circuit partitioning that would normally be done—input match, FET stage(s), and output match—enables the user to quickly confirm post-layout performance with schematic-based estimates from earlier in the design. Don’t forget to include the PDK’s bond pads as part of the schematic simulation, and if possible, the EXTRACT design.

As a case in point, Figure 10 shows a simple drain manifold added as a transition from the drain structure of the FET used in this example to a 100um-long section of 50um line on 50um-thick GaAs, using AWR APLAC® HB simulation with AWR AXIEM® 3D planar EM analysis.

2021 05 design 10

Figure 10. Drain manifold transition in layout (left) and comparison of pre- and post-EXTRACT simulation (right).

 

The resulting PAE and Pout simulation is also shown exhibiting the nonlinear performance degradation induced by these relatively minor but necessary features. In contrast, the bond pads themselves (Figure 11) offer very little change.

2021 05 design 11

Figure 11. Nonlinear performance of the ideal PA compared to adding three parallel bond pads at the PA output.

Step 6: Final Analyses

The final analysis step is where the design assumptions and simplification taken while creating the design can be revisited in the context of the whole design (now that it is seemingly complete). This step enables the designer to ensure that the whole is at least the sum of the parts and that in the process of focusing on parts of the design (i.e., partitioning into smaller pieces), two of the pieces were not inadvertently coupled together in such a way as to take away from the overall performance. From that perspective, it’s desirable to try and view this step as going up one level, so that the design parameters are the subblocks (input match, output match, FET/load, bias circuit, etc.) rather than the individual components within the subblocks. The performance criteria are the overall chip requirements and design closure is achieved when the performance criteria are met relative to what that analysis explores: electrical performance for EM, reliability for thermal, manufacturability for DRC, etc.

An analysis step is performed to ensure that second-order effects like EM coupling and thermal do not violate earlier design parametric constraints and assumptions. EM analysis will verify assumptions on source inductance and interconnect parasitics that can contribute to feedback paths, which may enhance instabilities. Although time consuming and requiring memory-hungry workstations, the greater the detail of the EM analysis, the greater the likelihood of finding potential oscillations or performance-starving parasitic effects. An EM simulation such as the AWR AXIEM analysis with the EXTRACT flow should be run at the top-level chip now rather than simply considering each design subblock separately. Going back and forth between the two is a great strategy for isolating any problems uncovered at this time.

Formalized finite-element method (FEM) thermal analysis reconfirms the channel operating temperatures. In the last decade or so, EM analysis has become robust enough to be included in the MMIC designer’s flow, and the same is now coming true with thermal analysis. Although new and different from the other substeps with which an electrical engineer may be familiar, thermal analysis is just too simple within the MMIC toolset and the payback too great NOT to do it. Underpinning many of the assumptions of the PA design is the operating temperature of the FET channel. With all the metallization in place after the layout is finalized, an electrothermal analysis can reaffirm decisions made about FET channel spacing and DC bias. The Cadence Celsius™ Thermal Solver integrated to AWR software can help close this loop in a matter of hours.

Should either the EM or thermal verification step fail by not achieving design closure, then interconnects can be made wider or shorter to minimize inductance or spaced further apart to avoid capacitance, or pHEMT fingers can be spaced further apart to relieve channel heating. In short, and without trivializing, for the GaAs pHEMT PA designer, the thermal consideration can in many cases be a secondary effect handled as an analysis step during verification. Of course, this is not withstanding aggressive thermal specifications or reliability requirements.

The concern in this step is that the designer may actually succeed in finding a problem with the design. Since the design parameters have been abstracted away, the designer runs the risk of not knowing what to fix (i.e., which interconnect to move, which bond wire to shorten, etc.) if the analysis does not create closure with the design requirements. The analysis tools will indicate that there is a problem, but without the ability to directly tie cause and effect through parametric models, the best guide is experience. Endless days or weeks of “move a line, run an EM simulation” rarely provide the answer. Last but certainly not least, design-to-manufacturing closure is needed: No design should be shipped without foundry-based DRC. AWR Microwave Office software includes both DRC and layout-versus-schematic (LVS) tools, and some foundries will do it for designers in a day or less.

Conclusion

Moving from one technology to another requires that certain skills and knowledge be transportable and transferable. The most basic of these skills is the effective use of EDA tools for designing the MMIC itself. In particular, PA designers need a strategy, design flow, and guidelines for how to start with specifications and a PDK and get to a point where the more compli- cated design requirements can be tackled.

In this white paper, the essential steps necessary for a typical PA design project have been illustrated using AWR Microwave Office software to design a relatively basic Class A GaAs pHEMT MMIC PA. The choice of a Class A amplifier was made to emphasize the flow itself and the need for designers to have a systematic approach to their design and to the design flow. It has been shown that at each step in the design flow, it’s important to clearly identify what it being designed, in terms of tying parameters to performance, and how designers will know when they are done with that step. Such an approach can be easily extended to more complicated classes of PAs and other circuit types.

References

1. M. Steer,  Microwave and RF Design: A System Approach,  SciTech Publishing,  2010.

2. S.A.Maas, The RF and Microwave Circuit Design Cookbook,  Artech House, 1998.

3. D. Wu and S. Boumaiza,  “Comprehensive First-Pass Design Methodology for High Efficiency Mode Power Amplifer, ” IEEE Microwave Magazine,  Vol 11. Issue 1,  pp 116-121. February 2010.

4. G. Gielen and R. Rutenbar,  “Computer-Aided Design of Analog and Mixed-Signal Integrated Circuits, ” Proceedings of the IEEE,  vol. 88,  no. 12,  pp. 1825–1854,  Dec. 2000.

5. S. Nuttinck,  B.K. Wagner,  B. Banerjee,  S Venkataraman,  Ed. Gebara,  J. Laskar,  H.M. Harrais,  “Thermal Analysis of AlGaN-GaN Power HFETs, ” IEEE Transactions on Microwave Theory and Techniques Vol. 51,  No. 12,  pp 2445-2452,  2003.

6. M. Alvaro,  A. Caddemi,  G. Crupi,  N. Donato,  “Temperature and bias investigation of self heating effect and threshold voltage shift in pHMET’s, ” Microelectronics Journal,  Vol. 36,  pp. 732-736,  2005.

7. S. C. Cripps,  RF Power Amplifiers for Wireless Communications,  2nd edition,  Artech House,  2006.

A. Platzker and W. Struble,  “Rigorous determination of the stability of linear n-node circuits from network determinants and the appropriate role of the stability factor K of their reduced two-ports, ” Third International Workshop on Integrated Nonlinear Microwave and Millimeterwave Circuits,  pp. 93-107,  1994.

Acknowledgement

Special thanks to Dr. Michael Heimlich,  a well-known member of the RF/microwave industry,  who joined AWR in 2001 and is today a professor at Macquarie University in Australia.

Search

CLICKABLES