Transistor Technologies for RFICs in Wireless Applications

Transistor Technologies for RFICs in Wireless Applications

This article presents a multidimensional technical analysis of RF transistor technologies for RFIC applications in cellular and personal communications service (PCS) markets. The first dimension describes the circuit design of each function block in the transmitter and receiver. The second dimension covers the suitability/strength of each transistor technology for each circuit design approach. The final dimension covers production uniformity and cost for the respective transistor technologies. GaAs heterojunction bipolar transistor (HBT) technology is used as the baseline for discussion and is compared with technologies that are currently available. This article concludes that GaAs HBTs combine the strength of both silicon (Si) bipolar transistors and GaAs FETs for RF applications.

Nan-Lei Larry Wang
EiC Corp.
Fremont, CA

Since the introduction of cellular telephone systems in the 1980s, the RF components market has grown steadily in terms of size and production volume. Cellular telephones represent the first high volume market for RF components and have transformed the industry. As a result, the need for high speed semiconductor components continues to increase.

At the beginning of the wireless era, the available technologies were Si bipolar CMOS, Si bipolar transistor and GaAs MESFET. Today, these methods are also the workhorse technologies. In recent years, other transistor technologies have been pursued as the operating frequency increased (a result of spectrum overcrowding and the associated demand for higher performance). As a result, GaAs HBT and high electron mobility transistor (HEMT) technologies are now being considered. More exotic technologies, such as transistors based on InP and SiGe or silicon on insulator, also are being explored. On the other end, CMOS in digital very large-scale integration is being studied for its performance suitability in the RF arena.

The HBT concept has existed for quite a while.¹ The first serious experimental study, which was based on the GaAs material system, was conducted more than 15 years ago.² Over the years, advances in epitaxial growth, device physics understanding and related know-how have made this technology practical. HBTs are used by major electronics companies throughout the world. Interestingly, HBTs based on other materials exhibit different performance features due to the individual material's characteristics. (Not all HBTs are created equal.) In this article, GaAs HBT technology is used as a baseline for discussion, and its pros and cons compared to various approaches are examined.

GaAs HBT: A Super-charged Bipolar Transistor
A GaAs HBT is a bipolar transistor that uses bandgap engineering technology.^1-3 Combined with the GaAs material properties, GaAs HBTs deliver improved performance over Si bipolar junction transistors (BJT) for several reasons. GaAs HBTs have much lower base sheet resistance, allowing large lithography dimensions. The devices also have no collector-substrate capacitance, providing a higher analog circuit gain bandwidth. In addition, GaAs has much higher electron mobility, allowing GaAs HBTs to achieve high f_T at a low V_ce range.

Bandgap engineering techniques improve HBT performance through the reduction of base sheet resistance. With a wide bandgap material in the emitter, the HBT base layer doping concentration can be much higher while a high current gain is maintained. The base sheet resistance value in a GaAs HBT can be as low as 200 W /sq, approximately an order lower than the resistance value of a typical Si bipolar transistor.

Since f_max = (f_T /8p R_b C_cb )^1/2 for a bipolar transistor, both R and C_cb must be minimized. To overcome the base sheet resistance limitation, the emitter width of a Si BJT must be reduced to minimize the base current crowding effect.³ Therefore, a fine lithography pitch must be utilized or a higher C_cb per emitter junction area will result.

A fringe benefit of a semi-insulating GaAs substrate is the removal of collector-substrate capacitance in the GaAs HBT. The collector-substrate capacitance is a major contributor to the delay time constant in an analog circuit,⁴ resulting in lower gain bandwidth. The electron mobility of GaAs is much higher than Si (8500 cm² /V-s of GaAs vs. 1500 cm² /V-s of Si).⁵ In a GaAs HBT, the f_T is highest at low V_ce (below 1 V), which also is beneficial for a low voltage, low power analog circuit. However, Si BJTs require a few volts of V_ce to achieve a high f_T . Therefore, a GaAs HBT essentially is a super-charged bipolar transistor. Other than the different V_be (1.4 V for GaAs HBT vs. 0.8 V for Si BJT), the analog circuit design approach for both transistors is identical. Figure 1 shows the minimum transistor feature size vs. speed (f_T and f_max ) for GaAs HBT, GaAs FET and Si BJT devices. The GaAs HBT has the most relaxed dimensional requirements and achieves the best performance.

Circuit Design Approaches Used in RFICs
Figure 2 shows a typical wireless system transmitter/receiver block diagram. The function blocks are listed in Table 1 . Each function block can be designed using different approaches. The commonly used transistor technology for each function block is also listed.

This article differentiates two design approaches. The analog circuit approach is composed of transistors, diodes, resistors and capacitors. There are no inductors. This approach is similar to the design of an operational amplifier. The RF/microwave circuit approach addresses the use of inductors to impedance match the transistor's capacitance. A typical example is the power amplifier.

Without inductors in the circuit, the transistor circuitry loses performance due to the mismatch of impedances. Therefore, the analog circuit approach requires a higher performance transistor to achieve similar circuit performance. However, the inductor integrated on chip occupies a large and expensive chip area and causes crosstalk with the other on-chip inductor. In addition, the inductor resonates with the transistor capacitance in a narrow frequency band (unless a multipole matching circuit is used, and it requires the use of multi-inductors), which limits the circuit for a particular application due to the narrow frequency response.

Another factor exists in this comparison. Since the RF/microwave approach requires the inductor to match the transistor capacitance, any variation of the transistor capacitance value will result in a performance change. However, the analog design operates below the corner frequency. Thus, even if the transistor capacitance variation causes the corner frequency to shift, the performance below the corner frequency essentially remains the same. With design margin, an analog circuit approach is more reproducible. Of course, the designer can combine both approaches for the optimization of an application.

A Comparison of Transistor Technologies

Transistors for Analog Circuits
The issues used to compare transistor technologies include frequency response, noise, linearity and layout (chip size). With regard to frequency response, the Si bipolar transistor is produced on a conducting substrate, and a substantial amount of collector-substrate capacitance Ccs exists. In an analog circuit, this capacitance presents a large time constant, which delays the circuit response,⁴ as shown in Figure 3 . In the f_T and f_max measurement of a single transistor, this capacitance is shorted out (or tuned out) and does not produce any effect. Therefore, a circuit composed of Si BJTs will be slower (reduced frequency response) than a circuit composed of GaAs HBTs of similar fT and fmax.

When comparing FETs to BJTs in speed, the transconductance gm must be examined. A FET has a lower gm than a bipolar. Since voltage gain A_v = g_m R_L , a lower g_m requires a higher R_L for the same gain. However, R_L also determines the time constant through the t = R_L C relationship. Therefore, a low g_m means slower speed. A BJT is inherently faster than a FET in analog circuit frequency response.

The results of a simulation using a divide-by-two and a cascaded two-stage amplifier as the testing vehicle for several transistor technologies are listed in Table 2 . The transistor models for circuit simulation are provided by various foundry vendors. In this case, the advantage of the GaAs HBT frequency response is obvious. Considering noise in analog circuits, feedback commonly is used to provide a broadband match to 50 W . In addition to the cumulative noise contributed by the transistors, the feedback resistors add noise to the circuit.

In multistage analog circuits, the voltage signal is amplified, whereas in RF/microwave circuits, the signal power is amplified. Therefore, the noise figure (NF) equation of the cascaded RF/microwave circuit does not apply directly in this case. However, the basic concept is the same: High gain in the first stage reduces the noise contribution from the following stages to the overall noise. The NF of an analog circuit is determined by the transistors and resistors as well as the circuit configuration.

Table 3 lists the contribution of noise in a simple resistive shunt feedback amplifier input stage. The feedback provides a 50 W input impedance to the amplifier. Even in the simplest feedback circuit, the resistors contribute a significant portion of noise. In linearity terms, the GaAs FET has many nonlinear sources, such as C_gs , g_m , C_gd and r_o . Generally, there is no dominant element in nonlinear behavior and the effect often cancels out.⁶ Therefore, balancing the nonlinear elements becomes critical in achieving good linearity. However, while good linearity is achievable in GaAs FETs, uniformity in production is a challenge.

The BJT has a dominant nonlinear source: the exponential base-emitter junction I-V curve. However, with large transconductance, the bipolar transistor can use the significant negative feedback from the emitter resistance⁷ to linearize the transistor. Since negative feedback is effective and reproducible, the bipolar transistor can be made linear in the manufacturing environment.

In the analog circuit layout, transistor size often is not the dominant factor; the resistors, capacitors and interconnections form the bulk of the required area. GaAs transistors are produced on a semi-insulating substrate and are true three-terminal devices in terms of layout. (The backgating effect in the GaAs FET technology will not be discussed in this article.) However, in Si, the conducting substrate demands a tub or isolation ring for each transistor and a fourth terminal connecting to the tub/ring is needed for every transistor. The tub terminal must be connected to ground or V_cc . Interconnection to the fourth terminal of the Si bipolar results in an active die size that is considerably larger than it would be if the same circuit was laid out on GaAs.

Transistors for RF/Microwave Circuits
Transistor technologies for typical RF/microwave circuits are compared with respect to noise, matching, bias scheme and breakdown voltage. As discussed in the analog circuit comparison, the linearity also applies for the RF circuit. Analog circuits cannot offer the lowest NF because of the required resistors. In low noise amplifier applications, the RF/microwave circuit approach is used.

The GaAs FET indisputably offers the lowest noise. However, for many wireless applications (such as cellular and PCS), the environment is interference dominated.⁸ The NF is only most important in satellite communications (such as Global Positioning System and direct-broadcast service) where background noise from the sky is dominant. Therefore, the bipolar transistor (Si or GaAs) can offer an acceptable NF in many interference-dominated applications.

In the case of matching, the input equivalent circuit of a GaAs FET can be considered as a capacitor. On the Smith chart, this impedance value corresponds to a high reflection coefficient and presents difficulty in achieving an easy match. In reality, the limited Q factor of the matching circuit eases the matching problem, but at the price of loss. The loss of the input matching circuit raises the NF of the entire circuit. Therefore, the GaAs FET low noise amplifier's NF is worse than the theoretical value of the FET. The input impedance of the BJT is more resistive and closer to the center of the Smith chart, making input matching easier.

On the output side, both the GaAs FET and BJT present a finite output conductance, resulting in easy output matching. The GaAs HBT has a very low output conductance and the |S22| at low frequencies is close to 1. Therefore, output matching can present a challenge for the GaAs HBT at low frequencies.

To compare bias schemes, the GaAs FET can be considered either in the enhancement or depletion modes; that is, the pinchoff voltage can be either positive or negative. A negative gate-to-source bias voltage is required for the depletion mode.

In a wireless application where pulse operation (burst, as it is referred to in wireless terminology) is required, the circuit must be powered down to save DC energy. Typically, the FET is turned off through a switching p-channel device along the drain line in the DC path. Both the negative gate bias and the p-channel device add cost and PCB space. For the bipolar in a common emitter configuration (Si or GaAs), the negative voltage charge pump is not required. In the burst application, the transistor is turned off when the base current is removed. Therefore, the Si or GaAs BJT is much simpler and cost effective.

Figure 4 shows breakdown voltage performance with the upper right-hand corner representing the best performance for RF power applications. The typical MESFET breakdown voltage is approximately 15 V (strongly process dependent). This voltage level is sufficient for a normal RF/microwave circuit in wireless operation up to 6 V bias. For a high performance Si bipolar, the BV_cbo is approximately 10 to 15 V and BV_ceo is only 3 to 5 V. This voltage level is too low for many RF circuits. The GaAs HBT has a much higher and controllable breakdown voltage than the Si bipolar; its BV_cbo is approximately 20 to 25 V, and a BV_ceo greater than 15 V is achieved easily. Therefore, the Si bipolar is suitable only for low voltage operation. Both GaAs FETs and HBTs can operate up to 6 V without any concern.

Comparison Results
Table 4 lists a comparison of the three transistor technologies. All data are based on product data sheets or measurements of off-the-shelf parts instead of quoting numbers from the most advanced research and development publication results. Therefore, the list is an objective comparison of today's transistor technologies.

The summary clearly explains why BJTs are more suitable for analog circuits. As a result, Si BJTs are widely used in small-signal RFIC designs. In large-signal (RF power) operation, the breakdown voltage of high performance Si BJTs is too low. Thus, GaAs MESFETs become the workhorse. Because GaAs HBTs encompass the strength of both technologies, they clearly are the most versatile technology.

Production Considerations
Yield is a key consideration in high volume production for wireless applications. High yield is achieved economically only after reproducibility and uniformity are established. Relying on screening to remove the out-of-specification parts is not a viable volume production approach.

The key challenge for the GaAs FET is controlling the channel thickness following the gate recess etching. Typically, the channel thickness is below 0.1 m m. A 10 percent variation in channel thickness of a uniform dopant concentration results in approximately a 20 percent change in the pinchoff voltage. Therefore, the gate recess etch is the most critical control item from wafer to wafer and lot to lot. The pseudomorphic HEMT has a very thin channel with a 200 Å (typ) spacing layer, making uniformity an even greater concern. For good performance even below 3 GHz, the industry trend is to have a gate length of approximately 0.5 m m, which requires the use of I-line stepper or e-beam lithography. Gate length variation also affects performance. The latest challenge is breakdown voltage control. Due to surface defects and the horizontal structure, MESFET breakdown voltage always presents a challenge to processing engineers.

The key challenge in reproducing Si BJTs involves several critical parameters, including base layer thickness, base layer total doping (Gummel number) and emitter dimension width. The current gain is a function of the Gummel number, which also is affected by the base layer thickness. Since the base often is manufactured by diffusion, current gain and performance are still of concern.

As discussed previously, the emitter finger width must be reduced to minimize the base-current crowding effect. The industry trend currently is at 0.5 m m and decreasing to 0.3 m m. At the same time, the emitter-to-emitter pitch must be reduced proportionally to maintain performance. Therefore, the lithography for RF Si BJTs is extremely challenging.

The key challenge in reproducing GaAs HBTs begins with the epitaxial wafer. With a growth rate of approximately 2 m m/hr, a thin layer down to 100 to 200 Å is common. GaAs HBT layer thickness figures less than 500 Å have not been published. Therefore, the layer thickness and doping concentration of the epiwafer for HBTs are controlled tightly.

With the low sheet resistance of the base layer (a result of the heterostructure), an emitter width of 2 m m or greater is used to over 20 GHz. Therefore, the lithography for HBTs is the most forgiving among the three technologies. In terms of the breakdown voltage, the HBT also has the tightest control as a result of the epiwafer and the vertical structure. Therefore, GaAs HBT technology is the most manufacturable process.

Conclusion
The performance features of the bipolar transistor make Si BJT the preferred technology for analog circuits. GaAs MESFET is the prevailing technology in RF power applications because of its high breakdown voltage and high f_max combination. Therefore, when selecting transistor technologies, the circuit requirements and transistor features should be matched carefully. In the case of RF analog circuits, the selection criterion is based on BJT vs. FET rather than Si vs. GaAs.

Because the GaAs HBT is a super-charged bipolar transistor, the RF analog circuit is easily extended beyond 3 GHz without using an on-chip inductor. The technology also has the high f_max and breakdown voltage required for RF power applications. Therefore, GaAs HBTs have all the strong features of both Si BJTs and GaAs FETs. Except for the RF switches, GaAs HBTs can serve all the RF function blocks in a typical RF transceiver.

References

1. H. Kroemer, "Heterojuncture Bipolar Transistors and Integrated Circuits," Proceedings IEEE, Vol. 70, 1982, p. 13.

2. P.M. Asbeck, M.C.F. Chang, J.A. Higgins, N.H. Sheng, G.J. Sullivan and K.C. Wang, "GaAlAs/GaAs Heterojunction Bipolar Transistors: Issues and Prospects for Application," IEEE Transactions Electron Devices, Vol. 36, No. 10, October 1989, p. 2032.

3. R.S. Muller and T.I. Kamin, Device Electronics for Integrated Circuits, Chapter 5, John Wiley & Sons, 1977.

4. P.R. Gray and R.B. Meyer, Analysis and Design of Analog Integrated Circuits, Chapter 7, John Wiley & Sons, 1977.

5. S.M. Sze, Physics of Semiconductor Devices, Appendix H, John Wiley & Sons, 1981.

6. J.A. Higgins and R.L. Kuvas, "Analysis and Improvement of Intermodulation Distortion in GaAs Power FETs," IEEE Transactions Microwave Theory Techniques, Vol. MTT-28, No. 1, January 1980, pp. 9-17.

7. N.L Wang, W.J. Ho and J.A. Higgins, "AlGaAs/GaAs HBT Linearity Characteristics," IEEE Transactions Microwave Theory and Techniques, Vol. 42, No. 10, October 1994, p. 1845.

8. W.C.Y. Lee, Mobile Cellular Telecommunications Systems, Chapter 2, McGraw-Hill, 1989.

Nan-Lei Larry Wang received his PhD degree from UC Berkeley and has more than 18 years of experience in the RF/microwave/mm-wave field with over 30 technical publications. He was the HBT task leader of the DARPA-sponsored MIMIC phase II team and has worked at Raytheon, Rockwell and Nippondenso. In addition, he has designed cellular telephone RF transceivers under IS-19 (AMPS) and RCR-27 (PDC) specifications. Wang is on the editorial board of the IEEE Transactions of Microwave Theory and Techniques and currently is engineering VP at EiC Corp.