The rapid development of handheld devices has necessitated the transmission and reception of high data-rate signals. Despite the developments in wireless communication technologies, the design of CMOS power amplifiers (PA) continues to pose challenges in the design of wireless transceivers because of the low breakdown voltage in CMOS devices, the no-substrate viahole in the CMOS process, and the low quality of the passive components. In addition, in WLAN applications, especially those meeting the IEEE 802.11n standard, a highly linear PA is required because of the high peak-to-average power ratio (PAPR) and the wide data bandwidth signal. Therefore, PAs are required to operate in a large back-off point from the in a saturation power region, resulting in decreased efficiency [1]. Hence, the design of WLAN PAs using CMOS technology is a challenging task.

Current communication standards require that PAs support dual or multi-bands. Therefore, the demand for dual-band PAs has increased. In 802.11n applications, 2 GHz and 5 GHz bands must be supported. In dual-band operation, the design of the output transformer is crucial because it must provide optimal impedance with low loss performance. However, no detailed design for the optimization of dual-band PA (2 GHz/5 GHz PA) output transformers has been reported. In wafer-level package (WLP) technology, an integrated circuit is used as the wafer level. WLP is essentially a true chip-scale package technology because the resulting package is almost the same size as the die [2]. Because WLP is intrinsically a chip-size package, its primary advantage is its small form factor. Another advantage is its low packaging cost compared to die-level packaging [3]. WLP technology also allows the chip to be reliably attached directly to a printed circuit board (PCB) [4]. However, unlike other packaging methods, it is difficult to design a dual-band PA in a compact die area using WLP technology because there are several limitations in the usage of WLP technology in RF PA designs. Without proper design considerations, the required die area is greatly increased because of the minimum distance rule between the balls and their large diameter. Although a dual-band PA using WLP technology was developed in [5], the considerations and details related to the design of a dual-band PA using WLP technology have not been addressed.

In this work, several design issues in a dual-band PA using WLP technology are addressed and a design method is proposed. In addition, a detailed design for the optimal impedance of dual-band PA (2 GHz/5 GHz PA) output transformers with low loss is presented. In the optimal dual-band operation presented here, the selection of the number of turns in the primary and secondary windings of the output transformers is investigated to provide optimal impedance. We design and implement a fully integrated dual-band CMOS PA for 802.11n applications using WLP technology. The fully integrated 2 GHz CMOS PA exhibits a 19.5-dBm output power and a PAE of 13.1% at the -28 dB error vector magnitude (EVM) point; the 5 GHz CMOS PA achieves an 18.1-dBm output power and a PAE of 8.9% at the -28 dB EVM point.

The output transformer is a major component affecting the overall performance of the PA. In designing a highly efficient linear PA, the selection of the optimal impedance is crucial. In the initial design stage, load-pull simulations are typically performed to find the optimal region of the input impedance in the transformers (Z_IN). Therefore, an output transformer is required to provide the corresponding optimal impedance region with low loss performance.

Fig. 1(a) depicts the layouts of 1:2 and 2:2 transformers. The equivalent model of a magnetic-coupled transformer is shown in Fig. 1(b). L₁ is the net inductance of the primary winding and L₂ is the net inductance of the secondary winding. R₁ and R₂ are the net resistances of the primary and secondary windings, respectively. Considering the effect of the voltage induced through mutual coupling,

where M represents the mutual inductance between the primary and secondary windings. Z_IN is the input impedance of the transformer, and R_load is the output impedance of the transformer. R_load is typically 50 Ω. From (2) and (3), Z_IN is given by

Because the R₂ and L₂ of the 1:2 transformer almost equal those of the 2:2 transformer, the differences in R₁ and L₁ determine the input impedance (R_IN) provided by the transformer. Because the R₁ and L₁ of the 1:2 transformer are lower than those of the 2:2 transformer, the 1:2 transformer provides a low R_IN compared to the 2:2 transformer. Fig. 2 illustrates the simulated R_IN and the minimum insertion loss. The minimum insertion loss, which is the inverse of the maximum available gain, was previously used to evaluate the passive loss performance [6]. The two transformers are the same size. A 30-μm metal width is used in the transformer design. Both transformers have similar minimum insertion losses. Fig. 2(a) shows that the R_IN of the 1:2 transformer is lower than that of the 2:2 transformer. The power generated by the device is inversely proportional to R_IN [7]. Therefore, the power generated by the 1:2 transformer is higher. In a 5 GHz PA operation, the passive loss is usually higher, and its operation has a larger impact on the device’s parasitic structure than the 2 GHz operation has. Thus, the power generated by the device should be higher to compensate for the effects of high frequency. In addition, the load-pull simulations validated that the optimum region of Z_IN in the 5 GHz PA output transformer was lower than that in the 2 GHz PA output transformer. Therefore, in this design, the 2:2 transformer was selected for the 2 GHz PA operation [8], and the 1:2 transformer was chosen for the 5 GHz PA operation.

The layouts of the output transformer in each band are shown in Fig. 3. The sizes of the transformers are optimized for each target frequency. The output transformers are designed with a 3.4-μm Cu metal layer. However, EM simulations, optimum metal width, spacing, and the size are selected for both 2 GHz and 5 GHz PA output transformers. The windings of the 2:2 transformer used in the 2 GHz PA are 30 μm wide and spaced at 10-μm intervals. The windings of the 1:2 transformer used in the 5 GHz PA are 15 μm wide and spaced at 5-μm intervals. Typically, for high target frequencies, the optimal size for maximizing passive efficiency is smaller in on-chip transformer designs. Increasing the size of the transformer increases the Q of the transformer at lower frequencies, while it can degrade the Q of the transformer at higher frequencies mainly because of the large coupling between the transformer and the lossy substrate. Therefore, the optimized size of the 5 GHz output transformer is much smaller than that of the 2 GHz output transformer. The size of the 2:2 transformer is 495 μm × 495 μm, whereas that of the 1:2 transformer is 235 μm × 235 μm. Fig. 4(a) shows the minimum insertion loss in the output transformers. With the proper section of transformer size, the peak of the minimum insertion loss is located on the target frequencies. When each transformer is plugged into the power stage circuit, input and output capacitors are added in parallel, as shown in Fig. 3, to provide the optimal performances of not only PAE and output power but also linearity. Through a careful process of the iteration of the total circuit simulation to maximize overall performance, the optimal Z_IN of the 2-GHz/5-GHz PA output transformers are selected, as shown in Fig. 4(b) and (c).

Among the most recent packaging solutions, WLP is the preferred method because the packaging size is small and the required cost is low. Careful considerations are required for a dualband PA design using WLP technology in order to minimize the required die area. To save the die area, a method placing an active device or passive transformers under the WLP balls can be considered. Placing active devices that operate at low frequency, such as digital circuits, under WLP balls may not cause any difference in performance. However, placing active devices or passive transformers that operate at 2 GHz or 5 GHz frequency ranges causes the performance degradations because of the electric and magnetic coupling between the active devices and the WLP balls. In particular, the passive transformer should be kept outside the area of the balls in order to minimize their impact. Because the increasing distance between the designed transformer and the balls also increases the required die area, the selection of the proper distance is importance to save the die area without causing the performance to degrade. Based on EM simulations, a minimum distance of 40-μm between the transformers and balls was chosen in this work.

Fig. 5 shows the schematic of the PAs. Both the 2 GHz and the 5 GHz PAs have differential amplifiers with a cascode structure. To achieve a high gain, the 2 GHz PA has two stages (i.e., the driver stage and the power stage), and the 5 GHz PA has three stages (i.e., driver stage 1, driver stage 2, and the power stage). An adaptive bias circuit is added to improve the EVM performance of the 2 GHz and 5 GHz PAs [9]. Fig. 6(a) shows a WLP structure with PCB and die. In the initial design stage, its equivalent ball model is used for total circuit simulations. Fig. 6(b) shows the equivalent WLP ball model. Because the diameter of ball is 130 μm, and the minimum distance between two balls is 300 μm in this work, the increased number of balls increases the required die area. Thus, the number of balls required in the dual-band PA design should be reduced by sharing. Fig. 7 shows a die photograph of the fabricated dual-band PA. As shown in Fig. 7, the power stages (2G PA) in the 2 GHz PA and the power stages (5G PA) in the 5 GHz PA share the PA VDD ball. In addition, the driver stage (2G DA) in the 2 GHz PA and the driver stage 1/driver stage 2 (5G DA1/DA2) in the 5 GHz PA share the DA VDD ball. In the driver stage design, the DA GND ball is also shared by 2G DA and 5G DA1/DA2. In this case, the performance is significantly compromised by the presence of connection routings. The parasitic effects in the connection routings and the balls are caused mostly by magnetic and capacitive coupling. Therefore, proper EM modeling is required to capture these couplings. Through full EM simulations, all transformers (input/inter-stage/output transformer) were re-modeled, including the surrounding balls and the closed signal/VDD/ground routings. All coupling effects are captured and included in the circuit simulations. Fig. 8 shows examples of a 2-GHz inter-stage/output-stage transformer, including the WLP balls and routings applied to capture interactions and parasitic effects.

The fully integrated 802.11n CMOS PA was fabricated using a standard process of 65 nm CMOS. The size of the chip, including the test pads, was 2.04 mm2, as shown in Fig. 7. During the measurement procedure, the losses in the PCB trace were carefully de-embedded, but the pad and bond wire losses were not de-embedded. Fig. 9(a) shows the measured gain and the PAE versus the output power at 2.45 GHz. The input signal was a single-tone continuous wave (CW) at 2.45 GHz. With a 3.3-V supply, the measurement results showed a 27.7 dB power gain. The saturated output power was approximately 26.3 dBm with a 32.9% PAE.

Fig. 9(b) shows the measured EVM and PAE versus the output power at 2.45 GHz using an MCS7 802.11n signal (64-QAM 65 Mbps with a 20-MHz BW and 9.6 dB PAPR). In the 802.11n standard, the specification for EVM is typically required to be below -28 dB. At the -25 dB EVM point, the output power was 20.3 dBm with a 15% PAE. At the -28 dB EVM point, the output power was 19.5 dBm with a 13.1% PAE. The measured output spectrum for the 2.45 GHz MCS7 802.11n signal is shown in Fig. 10. The output spectrum satisfied the mask specification at an output power of 19.5 dBm with a -28 dB EVM. Fig. 11(a) shows the measured gain and PAE versus the output power at 5.8 GHz. The input signal was a CW of 5.8 GHz. With a 3.3 V supply, the measured gain was approximately 40 dB. The saturated output power was 24.7 dBm with a 22.2% PAE. Fig. 11(b) shows the measured EVM and PAE versus the output power at 5.8 GHz using an MCS7 802.11n signal (64-QAM 65 Mbps with a 20-MHz BW and 9.6 dB PAPR). At the -25 dB EVM point, the output power was 18.8 dBm with a 9.8% PAE. At the -28 dB EVM point, the output power was 18.1 dBm with an 8.9% PAE. The measured output spectrum for a 5.8 GHz MCS7 802.11n signal is shown in Fig. 12. The output spectrum satisfied the mask specification at an output power of 18.1 dBm with a -28 dB EVM. Table 1 shows the comparison with a recently reported dualband 802.11b/g/a CMOS PA. The PAPR of the 802.11n MCS7 in HT20 was 0.5 dB higher than that of the 802.11g or the 802.11a [11]. Therefore, the PA that uses an 802.11n signal should operate more linearly than the PA that uses an 802.11g/a signal. Without applying predistortion, the performance of the implemented 802.11n CMOS PA was comparable to those reported in [5,10].

In this work, we developed a fully integrated dual-band CMOS PA for 802.11n WLAN applications using WLP technology. The output transformer structures were investigated and optimized for optimal dual-band operations. A 2:2 output transformer was chosen as the output transformer for the 2 GHz PA, and a 1:2 output transformer was chosen as the output transformer for the 5 GHz PA. The 2 GHz CMOS PA produced a saturated output power of 26.3 dBm with a 32.9% PAE. The 2 GHz PA achieved -28 dB EVM at an output power of 19.5 dBm with a 13.1% PAE. It also satisfied the mask requirements without the need to apply predistortion. The 5 GHz CMOS PA produced a saturated output power of 24.7 dBm with a 22.2% PAE. The 5 GHz PA achieved -28 dB EVM at an output power of 18.1 dBm with an 8.9% PAE. It also satisfied the mask requirements without the need to apply predistortion.