A 6 GHz Integrated High-Efficiency Class-F⁻¹ Power Amplifier in 65 nm CMOS Achieving 47.8% Peak PAE

Abstract

This paper reports a “single-transistor” Class-F⁻¹ power amplifier (PA) in 65 nm CMOS, which operates at the microwave center frequency of 6 GHz. The PA is loaded with a Class-F⁻¹ harmonic control network, employing a new “parasitic-aware” topology deduced using a novel iterative algorithm. A dual-purpose output matching network is designed, which not only serves the purpose of output impedance matching, but also reinforces the harmonic control of the Class-F⁻¹ harmonic network. This proposed PA yields a peak power-added efficiency (PAE) of 47.8%, which is one of the highest when compared to previously reported integrated microwave/millimeter-wave PAs in CMOS and SiGe technologies. The amplifier shows a saturated output power of 14.4 dBm along with an overall gain of 13.8 dB.

1. Introduction

The exponential rise in the demand for a high data rate in cellular and WLAN communications requires the current systems to demonstrate their capability to handle the upcoming throughput requirements. In order to deal with this challenge, shifting the systems to higher frequency bands in the spectrum is the only viable solution. A significant module in wireless systems is the power amplifier (PA). Integrated power amplifiers operable at microwave/millimeter-wave frequencies focusing 5G-communications are expected to demonstrate high efficiency and wideband operation within a small form factor. In a transceiver system, the power amplifier is the most dissipative module, which consumes power from the system’s battery. Hence, in order to extend the battery life in portable devices, it is imperative to design a power amplifier that demonstrates a high power efficiency [1,2,3,4,5]. Moreover, recently, the Federal Communications Commission (FCC) has also emphasized maximizing the battery life of the client device operating at the new unlicensed 6 GHz band (5.925 GHz to 7.125 GHz) [5]. Therefore, there is a crucial need for a 6 GHz power amplifier that exhibits high efficiency and thereby improves the battery life of the client device.

“Single-Transistor” PAs are desirable for 5G-mobile communications as they offer low-voltage operation, a small form factor, and a high power-added efficiency (PAE) if the PA is harmonically tuned. Moreover, the “single-transistor” approach incurs lower power consumption and, hence, naturally exhibits superior power efficiency over the multi-transistor approach (e.g., cascode or stacked, etc.). The term “single-transistor” means transistors are merged (shunted) in a parallel combination but are represented as one device in the circuit. “Single-transistor” integrated Class-F/F⁻¹ power amplifiers operable at “5G-millimeter-wave frequencies” have been investigated in recent years. However, an integrated “single-transistor” Class-F/F⁻¹ power amplifier operable at a “5G-microwave frequency” of 6 GHz has not been reported so far in the literature. A single-transistor millimeter-wave Class-F/F⁻¹ power amplifier in SiGe-BiCMOS was reported in [6]. This power amplifier employed a wave-shaping network connected at the collector node of the active device, which allows it to transform from Class-F⁻¹ to Class-F when the frequency shifts from 24 GHz to 31 GHz. The amplifier was operating at 2.2 V and delivered 40.7% PAE along with a gain of 10.3 dB and an output power of 17.1 dBm. Very recently, a 65 nm CMOS single-transistor Class-F power amplifier operable at 5G-millimeter-wave frequencies was reported in [7]. The 1.1 V driven power amplifier rendered 46.4% peak PAE, a power gain of 10 dB along with an output power of around 14.75 dBm. To the best of the author’s knowledge, this paper reports the first “single-transistor” integrated Class-F⁻¹ power amplifier operable at a “5G-microwave frequency” of 6 GHz. The amplifier is designed in 65 nm CMOS technology. The proposed design employs a “parasitic-aware” Class-F⁻¹ harmonic control network connected at the drain terminal of the NMOS device. There are several Class-F⁻¹ circuit topologies based on the LC network [6,7,8,9,10]; however, the proposed Class-F⁻¹ LC-network is a new topology deduced using a novel iterative algorithm. A dual-purpose output matching network is incorporated in the design, which reinforces the waveform-shaping capability of the Class-F⁻¹ harmonic network along with performing the typical task of output impedance matching. The proposed amplifier offers one of the highest PAEs (47.8%) compared to many other monolithic microwave/millimeter-wave CMOS and SiGe power amplifiers reported so far.

This paper is organized as follows. Section 2 describes the details of the Class-F⁻¹ amplifier’s circuit topology, which includes a harmonic control network and dual-purpose output matching circuit. Moreover, the section also explains in detail the iterative algorithm employed to extract the component values of the harmonic network. Measurement results and a comparison table are reported in Section 3. Finally, Section 4 concludes the work presented in this paper.

2. Single-Transistor Class-F⁻¹ Power Amplifier

Figure 1 shows the proposed Class-F⁻¹ PA using a “single-transistor” approach, implemented in 65 nm TSMC CMOS technology. The active device in the design employs an NMOS transistor M₁. The total width of the device (W_T) is 384 µm, which is divided into six power-cell units. Each power-cell unit contains 32 fingers (N_f), and each finger is 2 µm wide (W_f). The amplifier is biased, with a drain supply V_DD of 1.1 V and gate supply V_GG of 480 mV. A 5 Ω resistor (R_g) and a small inductor (L_s) are incorporated, respectively, at the gate and at the source of the transistor for the purpose of establishing unconditional stability for the power amplifier.

The drain terminal is loaded with a novel “parasitic-aware” Class-F⁻¹ harmonic control network, which harvests the parasitic capacitance C_P (246 fF) in its implementation. The parallel resonant tank (L₂, C₂) behaves as an inductor at the fundamental frequency f₀, and combines its inductance with L₁ to tune out the parasitic capacitance C_P. However, the tank circuit (L₃, C₁) resonates at f₀ to create an open circuit, and, hence, does not interrupt the fundamental frequency’s path to resonate out the C_P. At the second harmonic frequency 2f₀, the resonant tank (L₂, C₂) acts again as an inductor, which resonates out the parasitic capacitance C_P. However, the tank circuit (L₃, C₁) becomes overall capacitive at 2f₀ and constitutes a low impedance path to the ground, thereby discarding the presence of the inductor L₁ and leaving the second harmonic to resonate with the remaining circuit, whereas at 3f₀, the tank (L₂, C₂) becomes capacitive and constitutes a series LC resonant circuit with the inductor L₁ resulting in the formation of a short circuit path to the AC ground. A second short circuit path for the third harmonic is also established by the tank circuit (L₃, C₁), which behaves as an overall capacitor, thus dumping 3f₀ into ground. Hence, with the help of this passive frequency-steering the employed harmonic network accomplishes the requirements for Class-F⁻¹ loading.

In order to determine the values of the inductors (L₁, L₂) and the capacitor C₂, a novel “educated” and deterministic iterative technique is employed. It begins by inserting an incremental capacitance value, ΔC in (1) in each k^th iteration step of contiguous steps and then executing and validating a sequence of equations in a loop to finally extract the required reactive component values. The following equations are loaded in MATLAB in a routine and then the values of the components C₂, L_1, and L₂ are extracted instantly at the end of the iterative search.

$${\mathrm{C}}_2={\left[\Delta \mathrm{C}\right]}_{\mathrm{k}}+{\mathrm{C}}_{\mathrm{P}}$$

(1)

where [ΔC]_k is an incremental value in fF (such as 1fF, 3fF, 7fF, etc.) at the kth step, with k = 1, 2, 3….n. Additionally, in the equations below, f_1o, f_2o, and f_3o are the fundamental, 2nd and 3rd harmonics of the 6 GHz microwave center frequency, while f_0R, f_2R, and f_3R are the calculated resonance frequencies in each iteration. In addition, L_eq@2fo and X_Leq@2fo are, respectively, the required equivalent inductance and the corresponding inductive reactance of the parallel resonant tank (L₂, C₂) to tune out C_P at the 2nd harmonic. Further, X_Ceq@3fo and C_eq@3fo are, respectively, the equivalent capacitive reactance and the corresponding equivalent capacitance of the tank (L₂, C₂) at the 3rd harmonic. Finally, X_Leq@fo and L_eq@fo are, respectively, the equivalent inductive reactance and the corresponding equivalent inductance of the tank (L₂, C₂) at the fundamental frequency. The parasitic “C_P” thus provides the starting “educated” search reference for C₂, while L₁ and L₂ are found deterministically using C₂. It is thus different from any “brute force” algorithm, which may employ roughly estimated trial values. Now, for this proposed iterative technique:

$${\mathrm{L}}_{\mathrm{eq}@2\mathrm{fo}}=\frac{1}{{\left({2\mathsf{\pi }\mathrm{f}}_{2\mathrm{o}}\right)}^2{\mathrm{C}}_{\mathrm{P}}}$$

(2)

$${\mathrm{X}}_{\mathrm{Leq}@2\mathrm{fo}}{=2\mathsf{\pi }\mathrm{f}}_{2\mathrm{o}}{\mathrm{L}}_{\mathrm{eq}@2\mathrm{fo}}$$

(3)

$${\mathrm{X}}_{\mathrm{C}2@2\mathrm{fo}}=\frac{1}{{2\mathsf{\pi }\mathrm{f}}_{2\mathrm{o}}{\mathrm{C}}_2}$$

(4)

$${\mathrm{X}}_{\mathrm{L}2@2\mathrm{fo}}=\frac{\left({\mathrm{X}}_{\mathrm{Leq}@2\mathrm{fo}}{\mathrm{X}}_{\mathrm{C}2@2\mathrm{fo}}\right)}{\left({\mathrm{X}}_{\mathrm{C}2@2\mathrm{fo}}{+\mathrm{X}}_{\mathrm{Leq}@2\mathrm{fo}}\right)}$$

(5)

$${\mathrm{L}}_2=\frac{{\mathrm{X}}_{\mathrm{L}2@2\mathrm{fo}}}{{2\mathsf{\pi }\mathrm{f}}_{2\mathrm{o}}}$$

(6)

$${\mathrm{f}}_{2\mathrm{R}}=\frac{1}{2\mathsf{\pi }\sqrt{{\mathrm{L}}_{\mathrm{eq}@2\mathrm{fo}}{\mathrm{C}}_{\mathrm{P}}}}$$

(7)

$${\mathrm{X}}_{\mathrm{L}2@3\mathrm{fo}}{=2\mathsf{\pi }\mathrm{f}}_{3\mathrm{o}}{\mathrm{L}}_2$$

(8)

$${\mathrm{X}}_{\mathrm{C}2@3\mathrm{fo}}=\frac{1}{{2\mathsf{\pi }\mathrm{f}}_{3\mathrm{o}}{\mathrm{C}}_2}$$

(9)

$${\mathrm{X}}_{\mathrm{Ceq}@3\mathrm{fo}}=\frac{\left({\mathrm{X}}_{\mathrm{L}2@3\mathrm{fo}}{ \mathrm{X}}_{\mathrm{C}2@3\mathrm{fo}}\right)}{\left({\mathrm{X}}_{\mathrm{L}2@3\mathrm{fo}}{- \mathrm{X}}_{\mathrm{C}2@3\mathrm{fo}}\right)}$$

(10)

$${\mathrm{C}}_{\mathrm{eq}@3\mathrm{fo}}=\frac{1}{\left({2\mathsf{\pi }\mathrm{f}}_{3\mathrm{o}}{\mathrm{X}}_{\mathrm{Ceq}@3\mathrm{fo}}\right)}$$

(11)

$${\mathrm{L}}_1=\frac{1}{{\left({2\mathsf{\pi }\mathrm{f}}_{3\mathrm{o}}\right)}^2{\mathrm{C}}_{\mathrm{eq}@3\mathrm{fo}}}$$

(12)

$${\mathrm{f}}_{3\mathrm{R}}=\frac{1}{2\mathsf{\pi }\sqrt{{\mathrm{L}}_1{\mathrm{C}}_{\mathrm{eq}@3\mathrm{fo}}}}$$

(13)

$${\mathrm{X}}_{\mathrm{L}2@\mathrm{fo}}{=2\mathsf{\pi }\mathrm{f}}_{\mathrm{o}}{\mathrm{L}}_2$$

(14)

$${\mathrm{X}}_{\mathrm{C}2@\mathrm{fo}}=\frac{1}{{2\mathsf{\pi }\mathrm{f}}_{\mathrm{o}}{\mathrm{C}}_2}$$

(15)

$${\mathrm{X}}_{\mathrm{Leq}@\mathrm{fo}}=\frac{\left({\mathrm{X}}_{\mathrm{L}2@\mathrm{fo}}{ \mathrm{X}}_{\mathrm{C}2@\mathrm{fo}}\right)}{\left({\mathrm{X}}_{\mathrm{C}2@\mathrm{fo}}{- \mathrm{X}}_{\mathrm{L}2@\mathrm{fo}}\right)}$$

(16)

$${\mathrm{L}}_{\mathrm{eq}@\mathrm{fo}}=\frac{{\mathrm{X}}_{\mathrm{Leq}@\mathrm{fo}}}{{2\mathsf{\pi }\mathrm{f}}_{\mathrm{o}}}$$

(17)

$${\mathrm{L}}_{\mathrm{T}}{=\mathrm{L}}_1{+\mathrm{L}}_{\mathrm{eq}@\mathrm{fo}}$$

(18)

$${\mathrm{f}}_{0\mathrm{R}}=\frac{1}{2\mathsf{\pi }\sqrt{{\mathrm{L}}_{\mathrm{T}}{\mathrm{C}}_{\mathrm{P}}}}$$

(19)

Since the parameters C_P = 246 fF (captured through parasitic extraction incorporating the transistor’s parasitic capacitance (C_ds) plus parasitics associated with the layout interconnects), f_1o = 6 GHz, f_2o = 12 GHz, and f_3o = 18 GHz are known, they are used to extract the required component values by employing the above iterative technique.

It can be seen from the above equations and

Table 1. Comparison of results extracted by the iterative technique.

ΔC (fF)	C2 (fF)	L1 (nH)	L2 (pH)	f3R (GHz)	f2R (GHz)	f0R (GHz)
1	247	2.80	357	18	12	5.66
2	248	2.75	356	18	12	5.7
5	251	2.59	354	18	12	5.8
9	255	2.42	351	18	12	6.0

that the required resonance conditions at the 2nd and 3rd harmonics (f_2R and f_3R) are automatically in agreement, and satisfaction of the resonance condition at the fundamental, f_0R, is the validating parameter in this iterative component extraction method. For the first three iteration values of ΔC, the deduced reactive components resonate close to f_0R, but for the last iteration value of ΔC (=9fF), f_0R comes exactly to 6 GHz thereby completing the component search. It can be noted that the proposed algorithm starts from a specific reference value and quickly converges to the solution, as shown in the table. It is judiciously designed around a single parameter, i.e., a parasitic capacitance, and, hence, is different from other optimization algorithms, which perform a search based on arbitrary values of the various parameters.

Determining the component values of the tank circuit (L₃, C₁) constitutes a separate sub-problem and have been evaluated for resonance at f₀ and for equivalent capacitive AC-bypass to ground at 2f₀ and 3f₀. Hence, setting C₁ to a suitable value of 3 pF, L₃ can be easily deduced as follows:

$${\mathrm{L}}_3=\frac{1}{{\left({2\mathsf{\pi }\mathrm{f}}_{\mathrm{o}}\right)}^2{\mathrm{C}}_1}=234 \mathrm{pH}$$

(20)

For the purpose of satisfying the Class-F⁻¹ condition, the output-matching network also has to present an open-circuit to the second harmonic so that this harmonic would reflect back towards the drain of the active device. A simple output-impedance matching can be established by just using the two components, inductor L_M and capacitor C_M; however, in order to stop the second harmonic from propagating towards the load, a simple technique is employed. In this technique, the inductor L_M (1.165 nH) is judiciously decomposed into a 2f₀-tank circuit by employing the design equations below:

$${\mathrm{L}}_{\mathrm{R}1}=\frac{3}{4}{\mathrm{L}}_{\mathrm{M}}=873.75\mathrm{pH}$$

(21)

$${\mathrm{C}}_{\mathrm{R}1}=\frac{1}{{4\mathsf{\omega }}_0^2{\mathrm{L}}_{\mathrm{R}1}}={201.5}^{}\mathrm{fF}$$

(22)

The above-mentioned design equations can be used to convert any inductor into a 2f₀-tank-circuit. The tank-circuit (L_R1, C_R1) offers an open-circuit to the second-harmonic and behaves as the inductor L_M at the fundamental. Hence, the proposed output-matching network adds the second harmonic content at the NMOS drain terminal. This causes further refinement in the voltage waveform, resulting in a reduction in the current–voltage overlap and, hence, improving the power-added efficiency.

The proposed Class-F⁻¹ harmonic control network, along with the dual-purpose output-matching network connected at the drain of the NMOS transistor, performs the desired wave-shaping operation as shown in Figure 2. The voltage waveform approximates itself as a half-sinusoidal wave as the second harmonic content is dominant, whereas the current waveform is approximating itself as a square wave due to the higher third harmonic content (Figure 3). Load impedance observed from the drain terminal of the NMOS device (Figure 4) maintains a constant impedance close to the optimum value of 50 Ω at the fundamental frequency (f₀). The impedance trace depicts a high impedance at the second harmonic frequency (2f₀) and low impedance at the third harmonic frequency (3f₀). The input impedance matching is realized with the help of a DC-blocking capacitor, C_B1, as shown in Figure 1. The output load-pull simulations of the power amplifier offer around 52% peak PAE at 6 GHz, as indicated by the contours on the Smith chart in Figure 1.

3. Measurement Results

Figure 5 shows the photomicrograph of the implemented power amplifier employing the 65 nm TSMC CMOS process with a PA size of 753 µm × 927 µm. A Cascade Micro-Tech Summit 11K probe-station along with infinity-quad-probe (50 GHz RF and 100-µm pitch) and a Rohde & Schwarz ZVA50 Vector Network Analyzer employed for on-wafer testing, as shown in Figure 5. Standard SOLT calibration was performed for the purpose of on-wafer small-signal S-parameter measurements with the results shown in Figure 6. The measured S₁₁ and S₂₂ values are under −10 dB for a considerable range of frequencies of interest. The measured maximum S₂₁ was 14.1 dB along with a −3 dB range of 3.67–7.4 GHz, thus offering a 3.7 GHz wide small-signal bandwidth (BW_3dB = 67.4%). Stability analysis reveals that the amplifier’s stability factor (k) is greater than 1, as shown in Figure 7, ensuring unconditional stability of the proposed power amplifier.

Large signal measurements were conducted with the help of a power meter and power sensors. The measured results captured at the operating frequency of 6 GHz are shown in Figure 8. The amplifier exhibits a peak power-added efficiency (PAE) of 47.8%, saturated output power, P_SAT, of 14.4 dBm and, a power gain of 13.8 dB. The output power at the 1 dB compression point, OP_−1dB, is 12.9 dBm. Figure 9 shows the measured PAE traces from 5 GHz to 7 GHz with respect to the input power. P_OUT stays above 12 dBm from 3.5 GHz to 7.5 GHz, whereas PAE remains higher than 30% from 4 GHz to 7.5 GHz, as shown in Figure 10. Normalized AM–AM and AM–PM distortion curves are depicted in Figure 11. The measured AM–PM distortion at P_{in, −1dB} (6 dBm) is less than 2°, indicating good linearity performance.

Table 2. Comparison of recently reported CMOS and SiGe-BiCMOS PAs.

Referenced Work	Process Technology	Architecture	Amplifier’s Class	Freq. (GHz)	Gain [dB]	PSAT [dBm]	Peak PAE (%)	Supply Voltage (V)	Area on-Chip (mm2)
This Work	65 nm CMOS	Single Transistor	Class-F−1	6	13.8	14.4	47.8	1.1	0.69
[6]	130 nm SiGe	Single Transistor	Class-F−1/F	31	10.3	17.1	40.7	2.2	0.27
[7]	65 nm CMOS	Single Transistor	Class-F	29	10	14.75	46.4	1.1	0.12 *1
[11]	180 nm CMOS	Cascode Stages	Class-F−1	4.6	11.6	27.8	32	3.6	2.32
[12]	65 nm CMOS	Differential	Class E/F	6	N/A	25.6	34 *2	1.8	9.03
[13]	250 nm SiGe	Push-pull Stages	Class AB	6	12	24	24.7	1.8	2.08
[14]	65 nm CMOS	Differential	Class AB	6	23.6	22.4	28.4	3.3	0.89
[15]	65 nm CMOS	Differential Cascode	Class AB	6	19.4	18.9	27.9	3	0.73 *3

*¹ Area without pads (active area); *² drain efficiency; *³ circuit is not fully integrated (output matching network is off-chip); green shading: mmW PAs; blue shading: Single-Transistor.

compares the proposed PA with several previously reported PAs in CMOS and SiGe. It can be easily inferred from Table 2 that the proposed amplifier yields the highest peak PAE (47.8%) as compared to the referenced papers. According to the author’s knowledge, there are no “single-transistor” integrated microwave Class-F/F⁻¹ PAs that have been reported so far; however, two millimeter-wave Class-F/F⁻¹ PAs with similar “single-transistor architecture” have been reported, and these have been included in the table, although they are in a totally different RF spectrum compared to the other PAs in the table. The proposed PA offers a very high efficiency and operationwise reasonable output power. The saturated output power of 14.4 dBm resides within the designated power limits outlined by the FCC [5] for establishing 5G communications at 6 GHz. Moreover, this power level helps to extend the client device battery life as well. Table 2 also shows that when compared to other 6 GHz power amplifiers, the proposed PA offers low-voltage operation (1.1 V) and small form factor.

4. Conclusions

A highly power-efficient Class-F⁻¹ PA for 5G communications at 6 GHz, employing a “single-transistor” design implemented in 65nm CMOS, is proposed. The amplifier employs an iterative algorithm based parasitic-aware harmonic control network along with a dual-purpose output-matching network. The amplifier offers a low-cost, low-voltage driven, area-constrained, wideband, and a highly power-efficient PA for establishing 5G communications at 6 GHz.