Computer-aided-engineering (CAE) tools can be pricey, but it is still possible to design a low-noise amplifier at a commercial foundry using low-cost, open-source software simulators.
Designing a monolithic microwave integrated circuit (MMIC) implies million-dollar mask sets and expensive computer-aided-engineering (CAE) software. But for those on a budget, is it possible to be fiscally responsible and still create a MMIC? To explore the possibilities, a GaAs low-noise amplifier (LNA) was chosen as an example target design, since it can be created by means of a linear circuit simulator and an S-parameter file, including noise data, as supplied by the device foundry.
Although this example relies on a CAE program used by students at Johns Hopkins University (JHU), such a design can also be accomplished with a number of free or low-cost design tools. A separate opensource CAE tool, ICED from IC Editors, was used for MMIC layout and design verification. Of course, the design power of a full-featured suite of programs, such as the Advanced Design System (ADS) from Agilent Technologies or Microwave Office from Applied Wave Research (AWR Corp.), should not be easily dismissed. Still, it is feasible to accomplish an effective MMIC design with far less.
The example circuit is an LNA designed for use from 1800 to 2400 MHz, ideal for a variety of wireless applications. It is based on the TQPED 0.5-m pseudomorphic high-electron-mobility-transistor (pHEMT) process and design library from GaAs foundry TriQuint Semiconductor. The free ICED program was used for integrated-circuit (IC) layout, design rule checking (DRC), and layout-versus-schematic (LVS) checking. Simulations were performed using a simple linear simulator (GeeCAD) available to students at JHU and that runs on the student version of Matlab mathematical software. A number of CAE software suppliers are known for providing free or low-cost versions of their software for student use, including Ansoft, AWR, Agilent, and Sonnet Software.
An Internet search found a low-cost RF linear simulator known as LINC2 from Applied Computational Sciences with linear simulation capabilities for commercial use. In combination with this linear simulator, a simple DOS program provided by TriQuint was used to calculate lossy lumped element inductor models to improve on the initial simulations using ideal inductances. Noise figure data and optimum noise match (opt) for the PHEMT transistor were provided by TriQuint Semiconductor. For comparison, simulations were performed with several "high-end" tools, including ADS and Microwave Office used with design libraries from TriQuint Semiconductor for its 0.5-m TQPED GaAs pseudomorphic high-electron-mobilitytransistor (pHEMT) process. A twoand- one-half-dimension (2.5D) EM simulation using Sonnet software was also performed.
A simple design approach was taken starting with a linear S-parameter file for a 300-m enhancement-mode (E-mode) pHEMT from the TriQuint TQPED process. Bias was chosen as 3 V and 4.4 mA for low DC power consumption. Shunt and Series stabilizing resistors were added to the drain of the pHEMT using values tuned to provide unconditional stability at 1 GHz and above. The stabilized pHEMT was only conditionally stable below 1 GHz prior to adding the matching circuits.
Four simple input matching circuit topologies were tried, each containing four lumped elements (two capacitors and two inductors); only one of these initial designs provided unconditional stability below 1 GHz. The initial output matching circuit also used four lumped elements (two capacitors and two inductors) and was designed to conjugate match the combined input matching circuit along with the stabilized pHEMT. Then both the input and output matching circuits were modified to provide DC voltages to the gate and drain of the pHEMT. Large capacitors were added to provide a short circuit match at RF while decoupling the DC bias. The original simulations used the Generic computer-aideddesign (GeeCAD) program created by Dr. Lee Edwards and Sheng Cheng and used by students in the JHU RF & Microwave I and II design classes. This simple linear simulator employs a text-based netlist much like the original Touchstone program from EEsof, or the original Spice programs from the University of California at Berkeley. For graphic illustration of the amplifier design, the schematic diagram is included rather than the netlist.
After the initial ideal element design, the inductor program from TriQuint was used to calculate "lossy" spiral inductor models. Capacitors and thin-film resistors were treated as ideal elements while interconnect was ignored for all linear simulations. The Sonnet EM simulation includes all interconnects as well as unintended layout parasitics that could be missed in the linear simulations. A simple model for interconnection that could be easily added is to approximate a typical 10-m-wide microstrip trace on 100-m-thick GaAs substrate using the rule of 1 pH inductance per 1 m of microstrip trace length. Due to the small circuit size of approximately 1.2 x 0.8 mm, interconnects were ignored since at these frequencies any modeling errors introduced by interconnection parasitics should be minimal. After replacing the ideal inductors with "lossy" inductor models, the LNA design was re-tuned to optimize performance.
The layout was created using ICED and standard cells from the TriQuint TQPED library. Layouts were created for the capacitors, resistors, and spiral inductors by modifying existing standard cells. Bond pads were added to provide DC bias inputs and contact points for ground-signal- ground (GSG) probe-station measurements. For LVS checking, a netlist was generated manuallyfortunately, the design was simple. It would be easy to make a mistake in the netlist or in the layout that could cause a significant mismatch between the simulation and the layout. This kind of mistake can be missed by even the better CAE tools, so special care is required in verifying the layout and schematic diagram when using separate tools for simulation and layout.
A standard E-mode pHEMT device with dimensions of 6 x 50 m was chosen as the active device for the LNA design, with the added goal of minimizing power consumption. The E-mode pHEMT has more gain and slightly better noise figure than the depletion-mode (D-mode) pHEMT devices in the TriQuint TQPED GaAs process.1 Also, the positive E-mode gate threshold tends to simplify the DC bias for a single positive battery supply, although negative threshold D-mode pHEMTs can also be designed for a single positive supply by using a resistor in parallel with a large bypass capacitor at the source. Noise-figure data at 3 V and 4.4 mA bias as well as 3 V and 8 mA bias were used for the design and comparison of performance while maintaining low DC power consumption. A comparison of S-parameters of the PHEMT plotted on the Smith chart shows negligible differences for the two DC bias points (Fig. 1). A second plot shows good agreement between the linear S-parameter file and the non-linear model, version 3 of TriQuint's Own Model (TOM3), using Microwave Office at the 3 V and 4.5 mA DC bias setting (Figs. 2 and 3).
In trying to optimize overall performance, there are tradeoffs among stability, gain, noise figure, and return loss. A small 1-nH source inductor improved the input match with minimal penalty to the noise figure. Shunt and series stabilizing resistors were added to the drain (i.e., output) causing a small increase in noise figure, but their impact is minimal compared to stabilizing resistors at the LNA input. The stabilized pHEMT is unconditionally stable at 1 GHz and above, but only conditionally stable below 1 GHz prior to adding the matching circuits.
Table 1 shows the minimum noise figure and optimal match at 2.1 GHz for the pHEMT S-parameter file (1), the stabilized pHEMT with an ideal 1-nH source inductor (2), the stabilized pHEMT with the lossy lumped element inductor model (TriQuint inductor program) (3), and the stabilized pHEMT with a rectangular spiral inductor model in Microwave Office using the TQPED_MRIND2 element (4). As can be seen, stabilizing the pHEMT added about 0.3 dB to the noise figure and changed the optimal match point. As Table 1 shows, there is very little difference in the value of the optimal reflection coefficient ?pt = 0.6 at +27C) between an ideal source inductor (2) versus the two lossy inductor models (notes 3 and 4).
The LNA's input matching circuit was designed for optimal noise figure at roughly the middle of the 1.8-to-2- .4-GHz design band. After combining the input matching network with the stabilized pHEMT, the output matching circuit was designed for a conjugate match to obtain the best gain and output return loss over the 600-MHz operating bandwidth. Large capacitors connected to substrate ground viaholes and shunt inductors in both the input and output matching circuits were used to provide an RF ground while allowing a DC bias path to the gate and drains of the pHEMT. Schematic diagrams of the LNA were then created in ADS and Microwave Office (Figs. 4 and 5). The layout of the 1.2 x 0.8 mm GaAs MMIC was performed in ICED (Fig. 6). Figures 7, 8, 9, and 10 show the gain, stability, noise figure, and return loss for the GeeCAD, ADS, and Microwave Office simulations, respectively. Note the similar agreement between the simple linear simulation with lossy inductors versus Microwave Office and ADS simulation using the TriQuint TQPED libraries. All simulations reveal unconditional stability for the LNA design.
Contiue to next page.
The single-stage LNA shows 10 dB gain at midband with a good noise figure of about 1 dB. Table 2 provides a summary of the simulated and measured performance for the LNA from 1.8 to 2.4 GHz, with measured results in parentheses.
Any unexpected parasitic elements can be found by performing an EM simulation of the layout. If the layout is particularly dense, there may be unintentional coupling not reflected in the linear simulations. While the Sonnet Professional EM simulator is not inexpensive, an EM simulator can be a valuable addition to the design process. The LNA layout was imported into Sonnet with internal ports replacing the gate, drain, and source of the PHEMT device. Figure 11 shows the Sonnet layout used to generate S-parameters of the physical layout by means of a 2.5D EM field solver. The S-parameters from Sonnet are then combined with the pHEMT's S-parameters for comparison to the ADS, Microwave Office, and Gee- CAD simulations (Figs. 12 and 13). There does not appear to be any unexpected parasitics in the layout that were uncovered in this EM simulation. Careful layout and spacing of the elements along with verification of the MMIC layout are key factors in a successful design.
After the LNA was fabricated by TriQuint, it was probe tested at the Dorsey campus of JHU. The measured results compare favorably to simulations from GeeCAD, ADS, Microwave Office, and Sonnet. Figure 14 shows very similar gain response for all of the simulations compared to the actual measured data (the solid red trace) with the amplifier biased at 3.6 V and 6 mA. Return-loss results in Fig. 15 show good agreement between measured data and all simulations with some subtle differences in S22 values. Finally, a noise figure meter was used to measure the noise figure at a respectable value of 1.4 dB versus the simulated 1.2-dB value shown in Fig. 16. Since individual probe-testable pHEMT devices were included in the fabrication, a final comparison between measured S-parameters and simulations using measured pHEMT device data from the same wafer may explain the minor differences in output return loss.
In short, it is possible to design a GaAs MMIC LNA with a low-cost microwave simulator and opensource layout tool, achieving good agreement between simulation-based expectations and actual measured performance. While care should be taken when using separate programs for simulation and layout in contrast to an integrated tool, success is nonetheless possible. The addition of an EM simulator provides invaluable insights into parasitic circuit elements that may be inherent in a layout but remain undetected by a linear circuit simulator. As the example shows, it is possible to design MMICs on a low CAE budget with good results.
Many thanks to TriQuint Semiconductor for fabricating GaAs MMICs for JHU students since 1989. Support from the software vendors, particularly Applied Wave Research, Agilent, and Sonnet Software, is greatly appreciated, and especially support from our local AWR applications engineer, Gary Wray. The author continues to be inspired by former co-teacher of the MMIC design course for the first 15 years, Craig Moore, as well as other co-teachers and students in the JHU MMIC Design course. Thanks also to Dr. Edwards who created the concept for a MMIC design course at Johns Hopkins and recruited the author to co-teach the course. Dr. Edwards, with Sheng Cheng, also created the simple-to-learn GeeCAD program, which is used for teaching RF/microwave design at JHU.
1. John E. Penn, "Design a MMIC LNA with GaAs PHEMTs," Microwaves and RF, November 2006.