A Compact, Low-Power, Low-NF, Millimeter-Wave Cascode LNA With Magnetic Coupling Feedback in 22-nm FD-SOI CMOS for 5G Applications

This brief presents a 2-stage, low-power, millimeter-wave (mm-wave) LNA that aims at first to reduce its passive area through a transformer formed with inductors between the gate and source sides of the input transistor in a source-degenerated cascode. While reducing the passive area, the introduced transformer creates a magnetic feedback that further reduces the LNA’s noise figure (NF). We provide a theoretical analysis of this feedback-based noise reduction method and validate it with simulations and experimental results. The LNA has been fabricated in 22-nm FD-SOI CMOS and has a core area of only 0.09 mm2. The measurement results show that this design achieves a minimum 2.1 dB NF and a maximum 23.1 dB gain between 23.7–28.5 GHz at a power dissipation of only 5.6 mW from a 0.6V supply. Furthermore, despite using only a 0.6V supply voltage, IIP3 can still reach −16.5 dBm at 28 GHz. Targeting mm-wave 5G applications, the presented LNA is among the best-in-class with a sub-1V supply voltage.


I. INTRODUCTION
W ITH the recent advancements of wireless standards, the fifth-generation (5G) systems require higher carrier frequencies into the mm-wave region to support wider bandwidths. Such demands pose major challenges in obtaining a low noise figure (NF) for low-noise amplifiers (LNAs) as the minimum NF of a single MOS device is directly proportional to its operating frequency [1].
Many methods have been proposed in the literature, focusing on different noise optimization techniques. Some works are based on boosting the equivalent transconductance (G m ) of the circuit without substantially changing the bias voltages [2], [3], [4]. Among G m -boosting techniques, magnetic-coupling feedback using a transformer is the most preferred due to its competence in wideband scenarios. In this regard, for instance, [3] applies a transformer between the gate and source Manuscript  of a common-gate (CG) amplifier, as shown in Fig. 1(a), while [4] places a transformer between the gate and source of the cascode transistor, as depicted in Fig. 1(b). In addition to the G m -boosting techniques, some works have proposed the use of feedback via a capacitor [5], [6], [7]. Fig. 1(c) illustrates the general configuration and the feedback capacitor connected between the gate and source of the common-source (CS) transistor of a cascode LNA. Similarly, [8] suggests a different feedback with a transformer now between the gate and drain of a CS amplifier, as shown in Fig. 1(d). However, none of these techniques focuses on the silicon area optimization, i.e., reducing the area occupied by the passive components. In addition to the above prior-art techniques, some new noise cancellation methods have been proposed in [9], [10]. While these techniques can be very effective at RF frequencies, their This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ performance at mm-wave is limited by the degraded quality factor of the lumped components and parasitic phase shifts. This brief presents a 2-stage cascode amplifier for lowpower low-voltage mm-wave 5G applications with a noise cancellation technique based on magnetic-coupling feedback. In this design, the inductor in the input matching network (IMN) and the degeneration inductor connected to the source of the input-stage CS transistor are vertically integrated within the same die section to form a transformer. This significantly reduces the area occupied by the passive components. Furthermore, the resulting magnetic coupling between the source and gate of the CS transistor improves the noise performance. To further investigate this improvement, a detailed noise analysis of the presented technique is conducted. As a proof of concept, a two-stage cascode LNA using the presented technique is implemented in GlobalFoundries 22-nm FD-SOI CMOS. The core area of the fabricated LNA is only 0.09 mm 2 . The measurement results show that this LNA achieves a minimum of 2.1 dB NF, a maximum of 23.1 dB gain, 3-dB bandwidth between 23.7-28.5 GHz and an FoM of 19.7 while consuming only 5.6 mW from a 0.6-V supply. To the best of the authors' knowledge, at a sub-1 V supply voltage, these are the best results among LNA designs with similar topologies for mm-wave 5G applications.
The rest of this brief is organized as follows. Section II provides the analysis of the technique to illustrate its effectiveness. Measurement results are given in Section III, while conclusions are drawn in Section IV.

II. CASCODE LNA WITH MAGNETIC COUPLING FEEDBACK
The noise-cancellation technique demonstrated in this brief fundamentally relies on creating negative feedback between the source and gate of a source-degenerated cascode for the purpose of altering the impedance seen at the gate. This is accomplished through a transformer by vertically integrating the inductor in the IMN and the degeneration inductor in the same die section. In the literature, two similar works have been proposed [11], [12]. The former is based on simulation results only, whereas the latter aims to expand the bandwidth of a CS amplifier. Neither provides insight into the noise of the system.
In this section, we conduct a detailed noise analysis to examine the effect of transformer design on noise performance. Furthermore, we investigate the extend of NF improvement with the presented technique. Finally, we discuss the complete circuit implementation. Fig. 2 illustrates a simplified circuit representation of the presented cascode amplifier including the magnetic coupling feedback network. Based on this simplified circuit and the analysis presented in [13], the mean-squared output current noise of the source resistance, I 2 o,sn , and the thermal current noise of the channel under the gate, I 2 o,sd , can be calculated as

A. Analysis of the Cascode Amplifier With Feedback
where where M = k L g L s denotes the mutual inductance of the transformer with k being the magnetic coupling ratio between the branches. The resistive (thermal) noise, V 2 s,n , coming from the source resistance and the channel thermal noise, I 2 d,n , of the transistors are the main noise sources in a cascode amplifier. Moreover, T and γ represent the absolute temperature and the excess channel thermal noise coefficient, respectively. Lastly, g ms describes the source transconductance with g ms = ηg m in the saturation region. Note that the impact of the cascode transistor on the noise factor is ignored. This is because the noise contribution of the cascode transistor should be divided by the square of the input transistor's gain, which eventually becomes negligible in deep sub-micrometer CMOS [14]. Therefore, using (1) and (2), the noise factor of a source-degenerated cascode amplifier with magnetic coupling feedback can be derived as and, thus In (7), L g is strongly constrained for good impedance matching, whereas L s is constrained for good linearity. The demonstrated technique introduces an additional parameter, M, to this noise factor equation for the purpose of improving the noise but without degrading the matching and linearity. To observe the effect of the different pairs of k and L g -L s on the noise performance, Fig. 3 presents the theoretical NF obtained from (7). These results indicate that the frequency reaching the lowest NF can be shifted with different transformer configurations. To achieve a lower NF, k should be negative up to 40 GHz and positive at frequencies higher than 40 GHz. Considering the 28 GHz center frequency of the presented LNA, k should be set to a negative value for a lower NF.

B. Performance Analysis
This subsection aims to explore how the presented technique can provide an improvement in the noise performance of a cascode amplifier by comparing among the various transformer designs and with the case without a transformer. Fig. 4 plots the simulated minimum NF curves of a single-stage sourcedegenerated cascode amplifier. Note that all the inductor and transformer models were obtained from an electromagnetic simulator with full consideration of all the layout parasitics. From Fig. 4, it can be concluded that this technique always offers lower NF min than the case without transformer. Furthermore, NF min is approximately the same for all transformer designs. As a result, these examples clearly confirm that the magnetic coupling feedback helps to improve the noise.
The presented technique not only improves the NF [see (7)], but also provides better input matching for the cascode amplifier. In [11], it is demonstrated that increasing k of transformer windings between the gate and source of a cascode amplifier will improve the amplifier's input impedance matching. On the other hand, the simulation results of available gain shown later in Fig. 5(a) reveal that this technique does not provide any substantial improvement, which is as expected since G m here does not change. Moreover, since it does not introduce any additional non-linearity sources into the circuitry, it does not negatively affect the LNA's linearity.

C. Device Size Selection and Transformer Design
The transistor sizes were optimized via Cadence circuit simulations. It was observed that, at a center frequency of 28 GHz, 50 μm (1μm × 50) provides the lowest NF min with an acceptable available gain and so it was chosen as the transistor width in this design.
After the transistor-size adjustment, a suitable transformer can be designed to provide the magnetic coupling between the gate and source of the CS transistor of the input-stage cascode. Fig. 5 depicts the available gain and NF of a cascode amplifier for different transformer designs. All the inductors and transformers used in these simulations have real layouts. According to the results in Fig. 5(b), each transformer reaches almost the same lowest NF. The only difference is the frequency at which the minimum NF is obtained, as discussed earlier. However, the available gain across all the transformers is roughly the same. Still, the design with L g = 321.7 pH, L s = 735.1 pH, k = −0.9 provides slightly lower gain than others because higher inductance and k require longer coils which increase metal loss. After all, the transformer with L g = 141 pH, L s = 518.8 pH, k = −0.8 has the lowest NF at 28 GHz with an acceptable available gain. Therefore, this particular configuration was chosen for the final transformer design.
Determining the transformer design largely completes the first stage of the amplifier. To increase the overall gain, a second stage cascode is added, but in contrast to the first stage, it does not have a source-degeneration inductor. The two stages are connected via a decoupling capacitor, which also functions as a part of the inter-stage matching network. In the final cascode design, an additional capacitor is connected to the gate of the CG transistor rather than providing an RF short to allow for a voltage swing on the gate proportional to the swing of the corresponding source. Finally, a simple  LC matching network is implemented as an output matching network. For more details of the structure, Fig. 6 shows the complete schematic of the fabricated two-stage LNA, including the transformer, the matching networks, and bias circuitry.

III. MEASUREMENT RESULTS
The LNA was fabricated in GlobalFoundries 22-nm FD-SOI CMOS and its micrograph is shown in Fig. 6. The fabricated chip was measured using a Cascade Microtech Summit 9000 Analytical Probe Station and MPI T40A GSG-100 wafer probes. The gate voltage of the CS transistor in the first stage was provided through Agilent 11612BC07 bias tee. The chip consumes 5.6 mW from a 0.6 V supply.

A. Small-Signal Measurements
S parameters were measured with Agilent E8361A PNA, and on-wafer calibration was performed with an AC-2 calibration substrate. The comparison between the measured and simulated S parameters is shown in Fig. 7. The measurement results show that the fabricated LNA achieves a maximum gain of 23.1 dB at 25.7 GHz, with a 3-dB bandwidth between 23.7-28.5 GHz and a return loss better than 10 dB from 23.8 to 29.3 GHz.

B. Noise Measurements
NF of the fabricated chip was measured with a Y-factor method using R&S FSW50 spectrum analyzer and Keysight 346CK01 noise source. As shown in Fig. 8, the minimum NF is 2.1 dB at 26.8 GHz, whereas the maximum NF within the bandwidth range of 23.7-28.5 GHz is 2.37 dB.

C. Linearity Measurements
For the input-referred 1-dB compression point (iP 1dB ), the measurements were performed with the power sweep function of the Agilent E8361A PNA and at 28 GHz; iP 1dB was measured at −16.5 dBm. To further characterize the linearity, the input-referred third-order intercept point (IIP 3 ) was measured as −17.7 dBm at 28 GHz by applying two-tone signals with 100 MHz spacing at 28 GHz center frequency. Table I summarizes the performance of our mm-wave LNA and compares it with state-of-the-art publications. Thanks to the noise performance improvement achieved with the presented technique and the additional fully depleted layer under the active devices in the selected technology, this LNA attains a higher gain and lower NF at lower power consumption than the leading designs in bulk CMOS. This results in a superior FoM, which is the highest reported in the literature for mm-wave 5G applications with sub-1 V supplies. Among the 2-stage LNAs, only [19] exhibits slightly higher maximum gain (0.1 dB higher) but is powered from a higher supply voltage. In terms of the minimum NF, only [18] has better performance but still requires a much higher supply voltage. Finally, the fabricated LNA occupies a core area of merely 0.09 mm 2 , which is only slightly larger than [18]. On the other hand, it is even smaller than the 1-stage cascade LNAs proposed in [17], [20]. Overall, the presented design is among the best-in-class for mm-wave 5G applications.

IV. CONCLUSION
In this brief, we presented a compact low-power transformer-based 2-stage mm-wave cascode LNA for wireless 5G applications. The main motivation of this brief is to reduce the large passive area occupied by inductors by means of employing a transformer instead of two separate inductors at the gate and source of the source degenerated CS transistor of the first-stage cascode. In addition to reducing the passive area, the introduced transformer creates a magnetic feedback, which in turn improves the NF. The presented design was fabricated in 22-nm FD-SOI CMOS and its core area is only 0.09 mm 2 . The demonstrated magneticcoupling feedback technique is not specific to any process technology and thus can also be implemented in bulk CMOS. The measurement results show that with only 5.6 mW power dissipation from a 0.6 V supply, the presented LNA achieves a minimum of 2.1 dB NF, a maximum of 23.1 dB gain between 23.7-28.5 GHz, return loss better than 10 dB between 23.8-29.3 GHz and an FoM of 19.7. Considering only the LNA designs with a similar topology requiring sub-1 V supply voltage, our design features the lowest NF and the highest FoM to date.