A 19.6–39.4 GHz Broadband Low Noise Amplifier Based on Triple-Coupled Technique and T-Coil Network in 65-nm CMOS

This paper presents a differential 19.6–39.4 GHz broadband low-noise amplifier (LNA) in 65-nm CMOS technology. The LNA consists of two cascode stage and one common-source stage. To achieve a wide bandwidth and low average noise figure, inter-stage peak-gain distribution technique and transformer-based triple-coupled technique are developed. Besides, a new compact T-coil-based network is proposed to neutralize the parasitic capacitors and enlarge the gain. The measure results show that the 3-dB bandwidth is from 19.6 to 39.4 GHz, the maximum gain is 23.5 dB, and the noise figure (NF) is from 3.7 to 5.8 dB. The dc power comsumption is 46 mW with 1V supply voltage. The input P1dB is −17 dBm at 30 GHz.


Introduction
Recently, with the development of 5G millimeter wave communication and radar systems, several standards at the frequency bands of 24 GHz, 28 GHz and 39 GHz have been issued [1], which leads to a close watch on the wideband circuits and systems. LNA, as a critical component at the mm-wave wideband receiver, should achieve low noise figure, high gain and good gain flatness. Several efforts have been made to realize the wideband LNA in mm-wave frequency. To enhance the gain performance and lower the noise figure, G m -boost technique is proposed in common-gate LNA [2,3]. Gm-boost technique can be realized by active device, capacitive coupling and magnetically coupling. The magnetically coupling between the transformer coils greatly boosts the effective transconductance of the transistor, which can be realized in both single-ended and differential topologies [3]. A three-coil transformer should be used in the differential topology, realizing the input matching and Gm-boost function simultaneously. While, a parallel or serial small inductor is placed at the source of the common-gate transistor in cascode LNA [4,5]. Both techniques are suitable for single-end structure, which is sensitive to the power supply node.
To cover the various frequency bands allocated for 5G communication, broadband is also important. The broadband LNA is still an active research topic. Many efforts have been made to improve the bandwidth without sacrifacing the noise figure, gain, and power consumption. Single-peak staggering technique has been used in [6] to increase the bandwidth of a LNA. In the LNA, the resonant frequencies of each stage are evenly distributed, therefore creating an overall broadband total gain. However, it needs multiple resonant tanks which would consumes a lot of chip area. The single peak staggering technique also sacrificing the gain. Futhermore, since the noise figure is dominated by the first amplification stage, the single-peak staggering technique is not effective in extend the bandwidth of noise figure. Differential wideband LNAs are also introduced in [7], magnetically coupled resonators and Inter-stage transformer peak splitting techniques are used to achieve a wide bandwidth. But, it is suffered from the poor noise performance, because of the unproperly peak-gain distribution.
In this letter, we present a broadband LNA utilizing the peak-gain distribution technique and transformer-based triple-coupled technique to achieve the wide bandwidth and low average noise figure. A new compact T-coil-based network, placed at the source of common-gate transistor in the first cascode stage, is also developed to enhance the gain performance. The proposed LNA achieves a peak gain of 23.5 dB and minimum noise figure of 3.7 dB over a wide frequency range of 19. 6-39.4 GHz.
This paper is divided into four sections. Section 2 describe the proposed LNA, followed by the three main techniques, which are inter-stage peak-gain distribution, transformerbased triple-coupled technique, and compact T-coil-based network. Section 3 provides the measured and simulated results and compares them with other existing broadband LNA. Section 4 concludes this paper. Figure 1 shows the schematic of the proposed LNA. In order to achieve the high gain and wide bandwidth, 2-stage differential cascode structure and 1-stage differential common-source structure are adopted in this work. Cross-coupled drain-gate feedback capacitors are applied for each stage to increase the stability, gain and isolation simultaneously [8][9][10]. Magnetically coupled resonators (MCRs) are exploited in the first and second interstage matching network, to realize a wideband gain flatness and low average noise figure, by adjusting the coupling coefficients and properly splitting the peaks of the MCRs. A transformer-based triple-coupled topology is used in the input matching network to realize the G m -boost function in differential structure and a wideband input return loss as well. Besides, a new compact T-coil-based network is proposed at the first stage to resonate with the parasitic capacitors at the source of the common-gate transistors to enhance the gain of the first stage of LNA. The 3D EM model of all passive devices are also shown in Figure 1, respectively. They are designed by the top three metal layers of the process to minimize the insertion loss. The pink metal traces represent the top aluminum layer, while the green metal traces stand for the top copper layer.

Inter-Stage Peak-Gain Distribution
To achieve a wide bandwidth, transformer-based magnetically coupled resonators (MCRs) are usually adopted as the interstage coupling network in mmW amplifier design [7,10]. In the symmetrical MCR, Z 21 can be calculated as [10]: where ω 0 and Q are defined as Assuming 2Q 2 1 − k 2 , the two pole frequencies ω H and ω L are calculated as [10]:    Figure 2a shows the first interstage MCR1, R 1 and R 2 represent the output impedance of the first stage and the input impedance of second stage, C 1 and C 2 includes the parasitic capacitance and the additional capacitance to resonating with the L 1 and L 2 , and k 1 is the coupling coefficient of L 1 and L 2 . Figure 2b shows the simulated transimpedance responses Z 21 of MCR1 as the k 1 varies from 0.1 to 0.7. A larger k 1 value results in a greater pole frequency separation and thus a broader bandwidth, but a larger gain ripple. The coupling coefficient can be varied by changing the distance between the centers of primary and secondary coils. The simulated NF min of the first stage, connected with MCR1, is shown in Figure 2c. When k 1 increases from 0.1 to 0.7, the average NF min over the target bandwith decreases. According to the simulated Z 21 and NF min , the coupling coefficient k 1 should be large enough to split the two poles cover the whole target frequency band. And the transimpedance responses of MCR2 is shown in Figure 2d, which is to neutralize the gain ripple of MCR1. The coefficients of 0.65 and 0.35 are used for MCR1 and MCR2, and the total responses of MCR1 and MCR2 is also shown in Figure 2d. Figure 3a shows the 3D view of the proposed transformer-based triple-coupled network used for differential structure of LNA and boosting the transconductance (g m ).

Transformer-Based Triple-Coupled Technique
The pink metal traces represent the primary coil (L p ) located on the top aluminum layer, and the green traces represent the secondary coils (L s and L g ) located on the top copper layer. The voltages at the gate and source terminals are out of phase to enhance the voltage swing at the drain terminal by crossing the outputs of the two secondary coils. The effective transconductance G m can be calculated as:  G m = g m 1 + sC gs (sL g − sM pg + sL s + sM ps ) + g m (sL s + sM ps ) (6) The effective transconductance increase with the negative coupling inductance of −M pg due to the cross connection. While the NF is also reduced with the increased G m and can be calculated as [7]: Besides, the additional coupling coefficients (k ps and k sg ) between L p /L s and L s /L g helps to achieve a wideband input matching performance. The input return loss (S 11 ) is shown in Figure 3b, varying with k ps and k sg , and getting the best performance when k ps and k sg is around 0.5 to 0.6. The EM simulation results are shown in Figure 3c, the inductance of L p , L s , L g is 234 pH, 219 pH and 656 pH, respectively. And the Q factor of L p , L s , L g is 11.1, 21.6 and 19.1. The coupling coefficient between those inductors k ps , k pg and k sg is 0.52, 0.65 and 0.57, respectively.

Compact T-Coil-Based Network
The cascode structure is widely used in mmW LNA design, but the common-gate (CG) transistors will contribute considerable noise at the output and decrease the gain performance at the operation frequency. Thus, a small series inductor is placed between the common-source and common-gate transistors to resonate with the parasitic capacitance [5], shown in Figure 4a. An inductor placed in parallel with this parasitic capacitance is also remedy to the problem [4], shown in Figure 4b. Combining the ideas depicted in Figure 4a,b would result in the circuit shown in Figure 4c. The differential structure of the ideas is shown in Figure 4d, which can be realized by using a T-coil-based network, including two series inductors (L m1 ), two parallel inductors (L m2 ) and two coupling coefficient (k m ). The 3D view of the differential T-coil-based network, shown in Figure 4e, can be actualized by a multi-taps inductor. Because of the virtual ground at the differential structure, the inductors can be folded together to save the area. And the coupling coefficient can be controlled by the space between the turns. The coefficient between L m1 and L m2 could influence the gain of the cascode structure, and the simulated G max of the first stage of the proposed LNA varying with different k m is shown Figure 4f. The G max rises with the increasing of k m , but the bandwidth decreases. So, k m set to 0.35 in this design.

In
In In

Measurement Results
The broadband LNA was fabricated in 65-nm CMOS technology. Figure 5 shows the die photograph of the LNA, which occupies an area of 990 µm × 365 µm including pads and consumes 46 mA from a 1-V supply. In this design, the passive devices such as the transformers, output pads, and neutralization capacitors are simulated using an EM simulator. The transistors with local connections are RC extracted based on the foundryprovided transistor models and extraction rules. The passive devices and the RC extracted transistors are then combined in the circuit simulator. In the process of measurement, we use a high-frequency probe system to access the IN and OUT pins of the proposed LNA to evaluate the intrinsic performance of the chip, while the other DC pins such as the bias voltage and supply voltage are wirebonded to a PCB board.   Figure 6 shows the simulated and measured S parameters of the LNA. At the specific frequencies of 24 GHz, 28 GHz and 39 GHz for 5G mmW communication and radar systems, the measured gains are 21.8 dB, 22.1 dB and 21.8 dB, respectively. The measured 3-dB gain bandwidth is 19.8 GHz, which ranges from 19.6 to 39.4 GHz. The fractional bandwidth is beyond 67%. As shown in Figure 6, the measured input return loss (S 11 ) is less than −10 dB from 25 to 41 GHz, and the measured output return loss (S 22 ) is less than −10 dB from 22 to 36 GHz.  The linearity performance of the LNA is also shown in Figure 8, the measured IP1dB is −17 dBm at 30 GHz.

-15 -17
Sim. IP1dB=-15dBm Mea. IP1dB=-17dBm  Table 1 summarizes the performances of the proposed LNA and compares them with other recently reported broadband LNAs. To compare the overall performance of the mmW LNAs, a commonly used figure-of-merit (FoM) is adopted, which is defined as [5]:

Conclusions
The broadband LNA for 5G mmW applications is implemented in a 65-nm CMOS process. The peak-gain distribution technique and transformer-based triple-coupled technique have been developed to simultaneously achieve the wide bandwidth and low average noise figure. The new compact T-coil-based network placed at the first stage is also proposed to enhance the gain performance and lower the latter noise. The LNA achieves a peak gain of 23.5 dB and a 3-dB bandwidth of 19.8 GHz from 19.6 to 39.4 GHz. It also achieves a minimum NF of 3.7 dB at 26 GHz with 46-mW power consumption. It is evident that the proposed approaches are suitable for broadband LNA design, and the proposed LNA exhibiting great FoM compared with previous works.