Q- and V-band MMIC Low Noise Amplifiers

Technical Feature

Q- and V-band MMIC Low Noise Amplifiers

This article describes MMIC low noise amplifiers for millimeter-wave applications using 0.15 µm pHEMT technology. The design emphasis is on the active device model. The deficiency of conventional device models is identified. A distributed device model has been adapted to obtain the equivalent circuit of a pHEMT at millimeter-wave frequencies. Two single-ended, low noise amplifiers are designed using this model for Q-band (40 to 44 GHz) and V-band (58 to 65 GHz) applications. A Q-band, two-stage amplifier showed an average noise figure of 2.2 dB with an 18.3 dB average gain at 44 GHz. A V-band, two-stage amplifier showed an average noise figure of 2.9 dB with a 14.7 dB average gain at 65 GHz. The design technique and model employed provide good agreement between measured and predicted results. This work also presents state-of-the-art performance in terms of gain and noise figure, compared with published data.

Byung-Jun Jang, In-Bok Yom and Seong-Pal Lee
ETRI, Korea

Recently, there has been a growing demand for millimeter-wave systems, including wireless LANs, car radars and telecommunications. For the successful deployment of these services, highly reliable, low cost and reproducible MMIC low noise amplifiers are required.¹ InP-based pHEMT MMICs have emerged as one of the key technologies for millimeter-wave low noise applications. Nevertheless, cost, manufacturability and reproducibility issues have not yet been resolved. The higher maturity of GaAs-based pHEMT processes yields superior reliability, lower costs and higher reproducibility of the devices than InP-based HEMT^2,3 for millimeter-wave applications.

When designing millimeter-wave low noise amplifiers, there are many difficulties compared to low frequency MMICs, such as the inaccuracy of the active device models, the need of EM simulation, etc. First of all, accurate and verified active device models are needed. MMIC LNA designs at V-band often require smaller devices than the standard device provided, together with the design rules, by most GaAs foundries. Also, the operating frequency can be higher than the upper frequency guaranteed by the model.

At present, millimeter-wave, low noise amplifiers are normally designed using scaled and extrapolated small-signal and noise models of relatively large periphery devices. Because of the need to both scale to smaller size devices and extrapolate to higher frequencies, the resultant model invariably leads to several design iterations.⁴

Another method uses a sample approach. When the models of unknown devices are needed, their measured parameters are used. This approach can reduce the possibility of inaccuracy due to scaling and extrapolation, but is prone to give a misleading model, caused by the selection of a non-representative wafer that can degrade design robustness.⁵

In order to circumvent the scaling and extrapolation problems, and to predict small-signal and noise parameters accurately, a distributed device model recently originated by TRW⁶ was used. This model is somewhat different from conventional models, where only measured device S-parameters are used for modeling. Instead, this approach consists of a cascade of one-fingered elementary devices, representing the active area of the transistor, fed by lumped passive networks. Because this model is based on a long-term database of one-fingered elementary devices, an accurate simulation up to V-band is possible.^7,8

This article describes the design approach and performance of Q- and V-band MMIC low noise amplifiers using the distributed device model. A V-band two-stage amplifier showed a 2.9 dB average noise figure and a 14.7 dB average gain at 65 GHz. To the authors' knowledge, this is the best reported performance using GaAs-based MMICs operating at this frequency.

Fig. 1 HEMT device layout and distributed model.

HEMT Modeling

MMIC LNA designs at millimeter-wave frequencies often require smaller devices than those provided by most GaAs foundries. Usually large periphery devices, such as 80, 120 and 200 mm, are the standard devices provided by foundries. Also, foundries develop the HEMT models based on DC to 50 GHz measured S-parameters and up to 40 GHz measured noise parameters. Therefore, scaling techniques have to be used from the standard devices and extrapolation techniques for millimeter-wave applications. Scaling of a linear HEMT model as a function of gate width should be done using scaling equations provided with the model. Scaling equations are normally valid for gate widths from 75 to 125 percent of the original gate width.⁹

The TRW 0.15 mm gate length, four-finger, 40 mm gate width, pHEMTs are used in the V-band LNA design. A 40 mm device size was selected because it provided a good impedance range for matching circuit design. The bias condition is 2 V, 50 percent G_mpeak . Because this device cannot be scaled accurately from a standard device model based on a larger size device, the distributed model shown in Figure 1 was used.

This model looks more complex than other traditional models. To simulate the device exactly, this model uses one-fingered elementary devices and LIBRA™ for modeling the gate and drain distribution networks and the source air-bridges.

According to FET basic theory, it can be reasonably assumed that any non-scalable effect is mainly associated with the passive structure of the electron device, while the intrinsic region can practically be scaled in a proportional way.^7,8 The distributed model embeds these "one-fingered" elementary device models within equivalent circuit networks simulating the interconnect structure of the device layouts. Using these LIBRA models and four one-fingered elementary devices, it is possible to obtain an accurate model up to V-band. Figure 2 shows the measured S-parameters and distributed model results from 1 to 65 GHz. They agree very well across the entire frequency range except for the phases of S₁₁ and S₂₂ .

Fig. 2 Modeled vs. measured S-parameters from 1 to 65 GHz.

Low Noise Amplifier Circuit Design

With the distributed device model, a two-stage low noise amplifier has been designed using LIBRA in order to optimize gain and noise figure. The distributed device model described in the previous section was linked to the simulator and employed in the design. Amplifiers were designed at Q- and V-bands.

A photograph of the V-band amplifier is shown in Figure 3 . A single-ended architecture using a common-source configuration was chosen for low noise figure and moderate gain. The matching circuits were designed in the following order: input, interstage and output.

The input matching network was the most important because it determined the noise figure. A source inductive feedback was used for stabilization and to improve the input impedance of the device simultaneously.¹⁰ The interstage matching circuit was designed to transform the output impedance of the first-stage transistor to the optimum input impedance of the second-stage transistor. The optimum input impedance of the second-stage transistor was chosen by adjusting the length of source inductive feedback. The feedback values were chosen to ensure a low noise figure while giving a moderate gain. The coupler structure was designed to give a direct impedance transformation between the stages in order to reduce the inaccuracy in the DC blocking capacitor. The output matching circuit was designed to transform the output impedance of the second-stage transistor to 50 . In order to improve the matching bandwidth, open-circuited stubs were used.

The drain bias was applied through on-chip 10 resistors. Radial stubs were used in the bias circuits because of their broadband ground characteristics and low insertion loss. Distributed RC networks were used to ensure broadband stability without introducing significant noise in the pass-band.

For the V-band design, metal-insulator-metal (MIM) capacitors were avoided for critical RF matching and used only for DC biasing to minimize the sensitivity to process variations. For ease of on-wafer testing, the amplifiers were designed to operate with one common drain and one common gate bias.

Fig. 3 The V-band LNA MMIC.

An important focus in the design is the electromagnetic (EM) simulation because the layout is very densely integrated to reduce the MMIC size. Therefore, all the passive structures were analyzed via a full-wave EM analysis tool (SONNET™ software). Before EM simulation, the microstrip circuits are divided into known structures such as MLIN, and unknown structures for which EM simulation is needed. Using these block level EM simulation results, the layout is completed. Afterward, all the passive structures including MIM capacitors and via holes were analyzed using EM. Because EM simulation is a very time-consuming job, it is important to divide the circuit effectively. In this circuit, the circuit was divided into three sub-blocks. For example, the input matching circuit and feedback line below the first FET including the via hole were entirely EM-simulated using four-port circuits.

The LNA was fabricated on a 100 mm thick pHEMT wafer. The chip size of the V-band amplifier was 2.2 by 1.5 mm.

After completing the EM simulation and final layout, the yield performance of the designed amplifier was calculated. To explain yield reasonably, a long-term database of one-fingered elementary devices from 90 wafers was used. Inserting these values into the LIBRA simulator's discrete-value data file, a realistic yield variation can be calculated.

The Q-band low noise MMIC was designed using the same methodology, except for the matching network. A photograph of the Q-band LNA is shown in Figure 4 . A MIM capacitor was incorporated in the interstage matching circuit for DC block to reduce the chip size. The dimensions of the Q-band amplifier were 2.2 by 1.5 mm.

Fig. 4 The Q-band LNA MMIC.

Measured Results

The MMIC amplifiers were fabricated with a TRW 0.15 mm pHEMT process on 4 mil (100 µm) thick GaAs substrates. The small-signal S-parameters and noise parameters of the amplifiers were tested with on-wafer probing. The measured and simulated performances of the V-band amplifier are presented in Figure 5 . They compare very favorably over the entire frequency range. A yield of 60 percent was achieved, using a gain specification of 14 dB minimum and a noise figure of 3.5 dB maximum, from 58 to 65 GHz. The gain and noise variation is approximately 1dB from 58 to 65 GHz. Figure 6 shows the gain and noise distribution of amplifiers from four wafers at 65 GHz. At 65 GHz, the amplifiers show 2.9 dB average noise figure and 14.7 dB average gain. This is the best average gain and noise figure for GaAs-based MMIC LNAs in this frequency band, compared with the literature.

Fig. 5 Simulated and measured characteristics of V-band amplifiers.

Fig. 6 Small signal gain (a) and noise figure distribution (b) of V-band amplifiers at 65 GHz.

The simulated performance, using the distributed device model, was in good agreement with the measured data. As can be seen, the MMIC LNA shows a reasonable noise figure, in good agreement with predicted results. However, the gain is slightly higher than predicted. But the gain slope and gain variation are in good agreement with the predictions.

The measured and simulated performance of a Q-band amplifier is presented in Figure 7 . The Q-band amplifiers showed an average noise figure of 2.2 dB with an 18.3 dB associated average gain at 44 GHz. The measured and simulated results show good agreement except for a difference in the magnitude of the gain.

Fig. 7 Simulated and measured characteristics of Q-band amplifiers.

Conclusion

A design method for Q- and V-band low noise amplifiers, based on a distributed pHEMT model and exact EM simulation, has been demonstrated in this article. This model is successful in providing an equivalent circuit for a pHEMT that cannot be predicted from a conventional model. MMIC amplifiers based on the distributed pHEMT model were fabricated and evaluated. A Q-band amplifier exhibits an average gain of 18.3 dB and an average noise figure of 2.2 dB at 44 GHz. A V-band amplifier exhibits an average gain of 14.7 dB and an average noise figure of 2.9 dB at 65 GHz. These results indicate that the distributed pHEMT model is effective for the design of millimeter-wave MMIC low noise amplifiers.

Table 1 compares the published data for V-band GaAs-based MMIC LNA amplifiers. The amplifiers described in this article show excellent gain and noise figure. This work demonstrates that the 0.15 mm GaAs-based pHEMT technology can be successfully extended to produce low noise amplifiers required for millimeter-wave systems, provided that careful circuit design methodology and suitable device models are developed.

Acknowlegments

The authors wish to thank Roger Tsai and Mike Aust of TRW Co. for providing the distributed active device model used during the design of the LNAs and for useful discussions and helpful suggestions.

References
1. K.F. Lau, L. Liu and S. Dow, "Recent MMW Technology Development and its Military and Commercial Applications," IEEE Radio Frequency Integrated Symposium , 1998, pp. 87-90.
2. D.L. Deung, et al., "High Reliability Non-hermetic 0.15 mm GaAs Pseudomorphic HEMT MMIC Amplifiers," IEEE Radio Frequency Integrated Symposium , 1999, pp. 153-156.
3. A. Bessemoulin, et al., "Comparison of Coplanar 60 GHz Low Noise Amplifiers Based on a GaAs HEMT Technology," IEEE Microwave and Guided Wave Letters , Vol. 8, November 1998, pp. 396-398.
4. M.D. Dufault and A.K. Sharma, "Millimeter-wave HEMT Noise Models Verified through V-band," IEEE MTT-Symposium Digest , 1996, pp. 1321-1324.
5. M.King, et al., "A Product Engineering Exercise in 6-sigma Manufacturability: Redesign of a pHEMT Wide-band LNA," GaAs MANTECH Technical Digest , 1999, pp. 1-4.
6. R. Tsai, M. Nishimoto and R. Lai, "Forecasting Methods for MMIC RF Yield," GaAs MANTECH Technical Digest , 2000, pp. 113-116.
7. A. Cidronali, et al., "A New Approach to FET Model Scaling and MMIC Design Based on Electromagnetic Analysis," IEEE Transactions on Microwave Theory and Techniques , Vol. 47, June 1999, pp. 900-907.
8. T. Kuwabara, et al., "Accurate Analysis of Millimeter-wave MMIC Power Amplifier Using Distributed FET Model," IEEE MTT-Symposium Digest , 1999, pp. 161-164.
9. TRW, "HEMT Device and MMIC Foundry Design," 1998.
10. C. Pobanz, et al., "A High Gain, Low Power MMIC LNA for Ka-band Using InP HEMTs," IEEE Radio Frequency Integrated Symposium , 1999, pp. 149-152.
11. K. Maruhashi, et al., "A 60 GHz Band Low Noise HJFET Amplifier Module for Wireless LAN Applications," IEEE MTT-Symposium Digest , 1996, pp. 13-16.
12. R. Lai, et al., "A High Performance and Low DC Power V-band MMIC LNA Using 0.1 mm InGaAs/InAlAs/InP HEMT Technology," IEEE Microwave and Guided Wave Letters , Vol. 3, No. 12, December 1993.
13. L. Tran, et al., "High Performance, High Yield Millimeter-wave MMIC LNAs Using InP HEMTs," IEEE MTT-Symposium Digest , 1996, pp. 9-12.

Byung-Jun Jang received his BSEE degree in 1990, his MSEE degree in 1992 and his PhD in 1997, all from Yonsei University, South Korea. His previous work experience includes two years with LG Electronics designing RF modules for mobile communications. Currently, he works for ETRI, South Korea, as a research and development engineer in the field of active components for satellite payloads. His professional interests are Hybrid and MMIC circuit design. He may be contacted via e-mail at bjjang@etri.re.kr.

In-Bok Yom received his BS degree in electronics engineering from Hanyang University, South Korea. Since February 1990, he has been a senior research staff member for the Communications Satellite Development Center of ETRI. Currently, he is a team leader for the satellite RF component team. His research interests include MMIC design and satellite payload system design. He may be contacted via e-mail at ibypom@etri.re.kr.

Seong-Pal Lee is currently a project manager for the Communications Satellite Development Center of ETRI.