A Wideband Reconfigurable CMOS VGA Based on an Asymmetric Capacitor Technique with a Low Phase Variation

Abstract

This paper presents a wideband digitally controlled variable gain amplifier (VGA) with a reconfigurable gain tuning range and gain step in a 65 nm CMOS process. A unique asymmetric capacitor-based reconfigurable technique is proposed to extend the gain tuning range and realize gain step reconfiguration. An active neutralization topology based on a stackless transistor is utilized to compensate for the additional phase shift introduced by the gain tuning. Moreover, a current-type digital-to-analog converter (DAC) is also integrated for easier precise gain control. With the asymmetric capacitor varying from 1000 fF to 200 fF with a step of 400 fF, the proposed VGA achieves a 12.2/9.2/6.1 dB gain tuning range with a 0.4/0.3/0.2 dB gain resolution, respectively. At the maximum gain tuning range mode, the measured minimum root-mean-square (RMS) phase error is 1.7° at 23.4 GHz. At the finest gain step control mode, the RMS phase error measured across 20–30 GHz is lower than 1.9°. The tested result also shows the proposed VGA achieves a peak gain of 13 dB with a 3 dB bandwidth of 21.4–29 GHz, and the output 1 dB compression point (OP_1dB) is up to 8.6 dBm at 25 GHz.

1. Introduction

For the fifth-generation (5G) new radio (NR) phased array beamformers, the variable gain amplifier (VGA) is the key building block, which has attracted increasing attention from industrial and academic fields. The VGA is designed to serve both purposes. Firstly, it can effectively compensate for the different losses caused by the phase shifter during phase shifting [1,2]. Secondly, enough amplitude weighting can be provided for a phased array to achieve high sidelobe suppression [3,4,5,6,7,8,9]. For millimeter-wave (mm-wave) phase shifters (PSs), the 6 dB gain tuning range of the VGA is sufficient enough to cover the loss variation [2]. However, it is more desirable for the VGA to have a high gain resolution and low phase variation, to avoid introducing extra gain errors and degrading the phase resolution of PSs. For phased array systems, sidelobe suppression is very important, which directly determines the signal quality of the entire link. In order to achieve less than −30 dB sidelobe suppression, for a 16-element phased array, a range of gain tuning of about 12 dB is required, according to Taylor’s method [10]. Based on the two applications mentioned above, a new generation VGAs should have reconfigurable characteristics of gain tuning range and gain step to simultaneously serve both purposes, which can greatly increase the flexibility of phased array systems. To the authors’ best knowledge, so far, the VGA with reconfigurable gain tuning range and gain step size has not yet been reported.

Moreover, a VGA with low additional phase shift during gain tuning is also very important, which can greatly simplify the complexity of phased array calibration procedures [9]. To this end, various phase-invariant VGAs have been proposed [11,12,13,14,15], such as the study by [12], which achieves a 7.5 dB gain tuning range with <3.5° root-mean-square (RMS) phase error across 27–42 GHz by introducing interstage inductance. However, the designs mentioned above all adopt multiple-stack transistor structures, which suffer from more complex circuit topologies and higher supply voltage values, compared with stackless topologies under the same technology node and the normal supply voltage recommended by the vendor.

To address these issues, a new technique named as asymmetric capacitor-based reconfigurable technique is proposed to extend the gain tuning range and reconfigure the gain step. Based on a stackless transistor structure, an active neutralization technique is adopted to minimize the additional phase shift during gain tuning. Furthermore, for achieving accurate gain control to reduce gain error, the chip also integrates a high-resolution digital-to-analog converter (DAC).

2. Design and Analysis of VGA

Unlike the widely used current-steering (Figure 1a) and Gilbert-cell-based (Figure 1b) VGA structures [16,17,18,19,20,21], the proposed reconfigurable digitally controlled VGA used a stackless common source (CS) topology, as plotted in Figure 1c, which has advantages such as a simpler circuit structure and lower power supply. Figure 2a presents the full circuit schematic of the proposed reconfigurable VGA, and in both stages, a differential CS structure was used. Among them, the input stage is a variable gain stage to achieve gain tuning, and the output stage is a fixed gain stage to realize high output power. Furthermore, in order to more easily achieve accurate gain control and high robustness against the process and supply voltage and temperature variations (PVT) [16,17,18,19,20], the designed VGA used a digital control method. The control voltages V_a and V_b were generated by a 7-bit current-type DAC control circuit [22]. Additionally, to achieve wideband matching and compact layout, transformer-based high-order matching networks were employed.

2.1. Asymmetric Capacitor-Based Reconfigurable Technique

The core idea behind the proposed asymmetric capacitor-based reconfigurable technique was to connect asymmetric capacitors C_x and C_y in series on the gate nodes of transistors (M₁–M₄), respectively, as shown in Figure 2a. It should be pointed out that since the value of C_x is different from C_y, it is called asymmetric capacitors. Adjusting the values of asymmetrical capacitors provides another dimension of gain control, and hence, reconfigurable gain tuning range and gain step can be achieved. Figure 2b depicts the structure of the applied asymmetric capacitors. To achieve three configurations, the adjustable capacitor C_x was designed to be composed of three capacitors and two switching transistors, which are controlled by the bias voltage of Ctr1 and Ctr2. In addition, to ensure the reconfigurable effect, it is necessary to ensure that the designed capacitances of the asymmetric capacitors are as close as possible to the desired theoretical value. Based on this, both of the asymmetric capacitors used in this study were metal insulator metal (MIM) topology, because of its high resistance to process deviation. Meanwhile, C_x and C_y used similar capacitor arrays.

To investigate the reconfigurable mechanism, the core of the variable gain stage is shown separately, and its simplified schematic diagram is shown in Figure 3a. For further and more intuitive theoretical analysis, the corresponding half-side small-signal equivalent circuit is also established, as shown in Figure 3b. Based on Figure 3b, the voltage gain can be calculated by

$$G=\frac{Vout}{Vin+-Vin-}\approx \frac{A_1{g}_{m1}-A_2{g}_{m4}}{-(\frac{1}{Z_L}+\frac{1}{r_{o1}}+\frac{1}{r_{o4}}+j\omega C_{gd1}+j\omega C_{gd4}+j\omega C_{ds1}+j\omega C_{ds4})}$$

(1)

where g_m refers to the transconductances of transistors M₁ and M₄; the C_gd and C_ds describe the parasitic gate-to-drain capacitor and drain-to-source capacitor, respectively; the r_o represents channel output resistance; the Z_L is used to characterize the load impedance. In addition, A₁ and A₂ are obtained by

$$A_1=\frac{Cx(\frac{1}{r01}+j\omega C_{ds1}+j\omega C_{gd1})}{(Cx+C_{gs1}+C_{gd1})(\frac{1}{r01}+j\omega C_{ds1})+(Cx+C_{gs1})j\omega C_{gd1}+C_{gd1}{g}_{m1}}$$

(2)

$$A_2=\frac{Cy(\frac{1}{r04}+j\omega C_{ds4}+j\omega C_{gd4})}{(Cy+C_{gs4}+C_{gd4})(\frac{1}{r04}+j\omega C_{ds4})+(Cy+C_{gs4})j\omega C_{gd4}+C_{gd4}{g}_{m4}}$$

(3)

where g_m1 and g_m4 are biased by V_a and V_b, respectively, which are generated by a 7-bit DAC. According to (1), it can be observed that the voltage gain is proportional to the difference between V_a and V_b; that is, the greater the difference between the two, the higher the gain. Based on this, to achieve the gain tuning of the VGA, V_a should not be equal to V_b. Furthermore, when setting V_a less than V_b, from (1), the maximum gain tuning range ΔG_max can be derived as

$$\Delta G_{\max }=\frac{A_1{g}_{m1,\min }-A_2{g}_{m4,\max }}{A_1{g}_{m1,\max }-A_2{g}_{m4,\min }}$$

(4)

where g_m,min and g_m,max are the transconductances of transistors biased in minimum and maximum control voltages. It should be pointed out that since the amplifier gain is in decibels, the voltage gain should be converted into decibels, after which the logarithmic operation is performed. Based on this, the normal difference operation becomes division when placed in the logarithmic operation. Therefore, the final derivation of the gain tuning range appears as a ratio. In addition, it is worth mentioning that the C_gd is not ignored but neutralized based on the proposed topology in deriving Equations (1)–(4). The detailed proofs of Equations(1)–(4) are presented in Appendix A.

Then, based on (4), the gain resolution G_r of the VGA can be calculated as

$$G_{\mathrm{r}}=\frac{\Delta G_{\max }}{2^n}=\frac{1}{2^n}\cdot \frac{(A_1{g}_{m1,\min }-A_2{g}_{m4,\max })}{(A_1{g}_{m1,\max }-A_2{g}_{m4,\min })}$$

(5)

where n is the control bits of the DAC. According to (4) and (5), the proposed asymmetric capacitor-based reconfigurable technique provides a new method to configure ΔG_max and G_r by adjusting the coefficients A₁ and A₂. Meanwhile, in order to keep the maximum gain of the VGA, it is needed to set A₂ ≈ 1. As shown in Figure 4a,b, the conventional methods are only obtained ΔG_max and G_r by controlling transconductance g_m of the transistors, whereas the proposed technique achieves another dimension for ΔG_max and G_r control. When A₁ is varied from 1 to 0, ΔG_max and G_r will be reconfigured. When the asymmetric capacitors C_x and C_y are designed to be greater than 1000 fF across 20–30 GHz, it can be calculated from (2) and (3) that the values of the coefficients A₁ and A₂ are approximately equal to one. Conversely, when they are smaller than 1000 fF, the values of A₁ and A₂ will be less than one. Thus, the gain tuning range and gain step can be reconfigured, which is adjusted to capacitors C_x and C_y.

From what has been discussed above, the capacitance of C_x was adjusted by 2-bit switched-capacitor array, which could achieve 200/600/1000-fF; the C_y was designed to be a fixed value with a capacitance close to 1000 fF, so that the coefficient A₂ was approximately equal to one, as shown in Figure 3a. By configuring different C_x values, the simulated small-signal gains versus frequency are plotted in Figure 4c. Among them, in order to observe the reconfigurable effect more intuitively, only the maximum and minimum gain states are shown in the results. At the minimum gain control mode, the gain of the VGA changes greatly as C_x increases from 200 fF to 1000 fF, while the gain is almost the same at the maximum gain control mode. As a result, the reconfigurable gain tuning range and gain step can be realized. The simulation results shown in Figure 4c agree well with the theoretical analysis.

2.2. Phase Compensation Technique

For mm-wave VGAs, low additional phase shift during gain tuning is also a very important performance metric [9]. As mentioned before, many scholars have conducted extensive research to realize low phase variation during gain tuning. For VGAs with CS topologies, the parasitic capacitor C_gd has been discussed in detail [23] as the most important factor causing the phase variation. In order to eliminate C_gd, a method is widely used in amplifier design that introduces a positive feedback capacitor, which is called the capacitive cross-coupled neutralization (CCCN) technique [24]. This technique has the advantages of simple structure and obvious neutralization effect. However, its disadvantage is also obvious, that is, only good neutralization can be achieved in a narrow frequency band, as the positive feedback capacitor changes with frequency. This greatly limits the design of this technique in wideband VGA circuit design. To overcome this problem, the active neutralization technique in the previous study [23] was employed. The core idea of this technique is to replace the positive feedback capacitor in the conventional CCCN technique with an active transistor. Since the auxiliary transistors (M₃ and M₄) and the main transistors (M₁ and M₂) have the same size, the C_gd of the two can be guaranteed to be the same with frequency changes, thus achieving good neutralization in the wide frequency band.

Figure 5 plots the simulated maximum phase variation in the classic and proposed active neutralization-based CS VGAs across 20–30 GHz under the same gain adjustment range. Both simulations were conducted based on the same circuit configuration. Across 20–30 GHz, the maximum deviation of the phase variation in the proposed VGA with active neutralization technique was below 0.3° during gain tuning. In contrast, the maximum phase variation results for the two classic CS VGAs were relatively large. Among them, one VGA with conventional CCCN technique exhibited the maximum phase variation of 1.6°, while another VGA without any neutralization techniques showed 8.6° phase variation at 30 GHz. These results fully illustrate two points: (1) the proposed VGA with active neutralization technique can effectively eliminate C_gd and achieve low phase variation; (2) good phase compensation can be realized in the wide frequency band, which is very suitable for broadband VGA design.

3. Measurement Results

The proposed reconfigurable digitally controlled CS VGA was fabricated in a 65 nm CMOS process, and its die micrograph is presented in Figure 6. In the case of not including PAD, the core area of this chip was 0.758 mm × 0.23 mm. The two-stage VGA had a total power consumption of 98 mW with a 1 V supply, and the selection of 1 V supply voltage followed the vendor’s recommendation under the corresponding process node. It is worth mentioning that, to achieve active neutralization, the auxiliary pairs in the proposed structure were on, which consumed slightly extra power to maintain the same gain. Even so, the total power consumption of the variable gain stage with the DAC was only 29 mW (including auxiliary pairs). The VGA presented in this paper consumed relatively high power consumption, which is because it achieved large output power. If a system does not need such high output power, the bias voltage of the output stage can be decreased, and the total dc power consumption of the proposed VGA would be reduced accordingly.

The VGA gain was controlled by V_a and V_b, which were generated by the designed seven-bit DAC. The measured 32 states of gains under different control modes are shown in Figure 7a, Figure 8a and Figure 9a, respectively. It can be observed that the proposed VGA achieved a 12.2/9.2/6.1 dB gain tuning range, respectively, when the bias voltage of Ctr1 and Ctr2 were configured from 11 to 10 and then to 00 (C_x varying from 1000 to 200 fF, stepping 400 fF). Implementation of a 6.1 dB gain tuning range was aimed to insert loss compensation of PSs in phased array systems. Implementation of a 12.2 dB gain tuning range was intended for gain tuning in each element to suppress the sidelobes. As for the intermediate state, it was a compromise that was reserved according to actual design requirements. Meanwhile, the peak gain remained basically constant with the changes in C_x. The measured 3 dB bandwidth was as wide as 7.6 GHz, from 21.4 to 29 GHz, with an approximately 13 dB peak gain.

To further achieve the ultrahigh average sidelobe suppression ratio of the beamforming to ensure beamforming quality and chain data rate, the VGA was suggested to achieve a gain step of less than 0.5 dB [10]. As for insertion loss compensation of PSs, higher gain steps were also required, to avoid introducing extra gain errors. Based on the above considerations, the proposed VGA was designed to have a 0.4/0.3/0.2 dB gain resolution, respectively. Achieving such a high-precision gain step control, it was necessary to design a high-precision DAC. According to (5) and the measured gain tuning range, the calculation results of the three gain resolutions are as follows:

$$G_{r\max }=\frac{\Delta G_{\max }}{2^n}=\frac{G_{\max }-G_{\min }}{2^6}=\frac{12.75-0.6}{2^6}\approx 0.4$$

(6)

$$G_{r\mathrm{mean}}=\frac{\Delta G_{\max }}{2^n}=\frac{G_{\max }-G_{\min }}{2^6}=\frac{12.9-3.7}{2^6}\approx 0.3$$

(7)

$$G_{r\min }=\frac{\Delta G_{\max }}{2^n}=\frac{G_{\max }-G_{\min }}{2^6}=\frac{13-6.9}{2^6}\approx 0.2$$

(8)

Furthermore, the three configurations for the min, mean, and max gain step versus the six-bit binary control word overall frequency band are plotted in Figure 7b, Figure 8b and Figure 9b, respectively. The proposed VGA basically realized three reconfigurable gain steps of 0.4/0.3/0.2 dB, respectively. Then, the measured input return loss (S11) and output return loss (S22) of the VGA for the different configurations are plotted in Figure 7c, Figure 8c and Figure 9c, respectively. It can be seen from the results that changing the capacitance of C_x will slightly affect the offset of S11, which, due to the C_x, will slightly change the imaginary part of the source impedance. Since the output stage was a fixed gain stage, S22 basically remained unchanged versus 32 control words. In addition, it is worth mentioning that, to achieve wideband matching and compact layout, the transformer-based high-order matching networks were adopted. By adjusting the coupling coefficient k of the primary and secondary coils, the values of resonance frequency (ω_L and ω_H) could be adjusted, thereby realizing the adjustment of the bandwidth [25]. Figure 10 presents the measured small-signal gains versus 32 gain modes at 25.4 GHz, which verifies that the proposed VGA achieves reconfigurable gain tuning range and gain resolution, in addition to realizing linear gain control.

The RMS phase error was used to characterize the phase variation in the VGA. The specific formula for calculating RMS phase error is expressed as

$$RMSPhaseError=\sqrt{\frac{1}{N}{\displaystyle \sum _{i=1}^N{\left|{\theta }_i-{\theta }_{ave}\right|}^2}}$$

(9)

where ϴ_i is the phase at the i state, and ϴ_ave is the average of the phases of all states. In this paper, the phase was extracted from the argument of S21. The RMS phase errors for all gain states were measured by taking the maximum gain state as a reference, and they are shown in Figure 11a. At the maximum gain tuning range mode, the measured RMS phase error was 1.7° at 23.4 GHz. Across 20–30 GHz, the measured RMS phase error was less than 5.5°. At the medium gain tuning range mode (0.3 dB gain step), the measured minimum RMS phase error was 0.5° at 25.2 GHz. Across 20–30 GHz, the measured RMS phase error was less than 2.4°. At the finest gain step control condition, the RMS phase error measured across 20–30 GHz was lower than 1.9°. At 30 GHz, a minimum phase error of 0.22° was achieved. Meanwhile, under the maximum gain state, the OP_1dB was up to 8.6 dBm at 25 GHz, as plotted in Figure 11b.

Table 1. Performance summary and comparison of the VGAs.

	This Study	ISSCC 2017 [11]	RFIC 2018 [12]	MWCL 2019 [14]	TMTT 2021 [15]
Technology	65 nm CMOS	40 nm CMOS	65 nm CMOS	65 nm CMOS	55 nm CMOS
Topology	Stackless	Two-stack	Two-stack	Two-stack	Two-stack
Supply (V)	1	1.1	1.2	1.2	1.3
Freq (GHz)	21.4–29	26–36	27–42	30–34.5	6.5–12
S11 (dB)	−9~−5	−20~−10	−30~−4	−28~−11	−14~−36
S22 (dB)	−17~−13	−7~−6	−50~−5	−30~−10	−14~−35
Peak gain (dB)	13	22.4	9.6	20.8	20.7
ΔG (dB)	12.2/9.2/6.1	8	7.5	10.6	18
Gain resolution (dB)	0.4/0.3/0.2	1	0.5	20.1	N/A
RMS phase error (°)	<1.9/2.4/5.5	<6 *	<3.5	<8 **	<4.5
OP1 dB (dBm)	8.6	13.7	2.5	−0.6 #	7.5 #
PDC (mW)	98	30.3	15.6	26.7	75
Core Area (mm2)	0.174	0.23	0.083	0.2	0.98

* Estimated from measurement results; ** phase variation; ^# calculated from IP_1dB with gain.

summarizes the performance of this research and compares it to prior studies. The proposed VGA is the only one that can reconstruct gain tuning range and gain resolution, which will greatly increase the flexibility of phased array systems. Additionally, it is the only one that adopts stackless transistor topology, thus achieving the lowest supply voltage, compared with other state-of-the-art multiple-stack transistor structures.

4. Conclusions

A 21.4–29 GHz reconfigurable digitally controlled VGA based on stackless CS structure with novel asymmetric capacitor-based reconfigurable and active neutralization phase compensation techniques was introduced. The proposed VGA achieved a 12.2/9.2/6.1 dB gain tuning range with a 0.4/0.3/0.2 dB gain step while maintaining the peak gain constant. At the finest gain step mode, the measured RMS phase error was <1.9° across 20–30 GHz. At 30 GHz, a minimum phase error of 0.22° was achieved. At the maximum gain state, it achieved a measured 13 dB peak gain and 8.6 dBm OP_1dB. The measurement characteristics demonstrate the proposed VGA is suitable for 5G mm-wave NR phased array beamformers that require reconfigurable gain tuning range and gain step with low phase variation.