C. Effects of Output Impedance
When the device width is increased both the output conductance and the drain current are scaled by the same factor. Therefore, the ratio of the output impedance (rds) to the optimum output impedance (ROPT) remains constant and no advantage can be gained in terms of drain efficiency. The only way to improve rds is by increasing the channel length; however this is not an option since it reduces the fT. Figure 7 shows the effect on the power added efficiency (PAE) asrds become comparable to ROPT. Asrds decreases, the output saturates at a lower power since a more significant part of the RF power is dissipated internally. To improve linearity the auxiliary amplifier must be biased to turn-on at a lower input power. However this moves the auxiliary amplifier closer to Class B operation reducing the efficiency even further. Also, a lowrds reduces the effectiveness of the active load-pull because less current makes it to the load. This effect can be mitigated by increasing the width of the peak device.
D. Effects of Breakdown Voltage
Thin oxide devices must use lower supply voltages for reliability reasons. Figure 8 shows how the supply voltage affects the PAEand PSAT. ROPT is reduced with the supply voltage to accommodate the lower voltage swing.
III. PROPOSED TOPOLOGY AND RESULTS The trends from section II are used to implement a reconfigurable DPA in a commercially available 90-nm CMOS process at 71 to 76 GHz. The reconfigurable option is bias controlled and allows the DPA to be transformed into a balanced PA in cases where low PAPR modulation schemes are used. The advantages of balanced PAs are improved gain, linearity, and stability, as well as added redundancy. To allow the option, symmetry between the two parallel stages is needed. Therefore, the main and the peak amplifier branches are designed using the same PA block as shown in Figure 9. The main PA is biased in Class AB instead of Class B to improve gain at mm-wave frequencies.
The PA transistor is sized for optimum fMAX and rds. An electromagnetic design approach with Agilent Momentum [9] is used for the verification of the passives including transistor feed networks which are shown in Figure 10 (b). Coplanar waveguide (CPW) based passives are employed to reduce losses in the silicon substrate [10]. The power splitter between the main and the peak state is implemented with a 90-degree hybrid coupler. The coupler is meandered and lightly loaded with via bridges to reduce the layout area. The EM simulated bandwidth of the coupler is found to be more than adequate to cover the whole 71-76 GHz range. All the passive networks are implemented on an ultra-thick metal layer available in the process. Feasibility of the circuit is improved by reducing the dependence of transistor parasitics by utilizing low qualityfactor (Q) matching networks. The final DPA layout has a total footprint of 1590 μm by 965μm. The layout including testing structures is shown in Figure 10.
In the DPA configuration the circuit produces 11.7 dBm of output power at 1-dB compression point (P1dB) with 30.6 % PAE. The balanced configuration has similar P1dB and PAE with 11.9 dBm and 30.1 % respectively. However, at 6-dB back-off from P1dB the DPA PAE is improved by 8 %. The improvement in gain of the balanced configuration is 3.2 dB.
The RF transistors in the 90-nm CMOS process used for this implementation have output impedances of about four times larger than ROPT, a knee voltage about 30 % of VDS, and are powered with 1.5 V supplies. Using the proposed Doherty PA model, Figure 11 shows that the predicted trends are within 5 % of the actual results.
IV. CONCLUSION
The feasibility of implementing a high efficiency Doherty amplifier in commercial CMOS processes has been discussed here. An adapted version of the Curtice-Ettenberg model was used to investigate design trends and limitations of the CMOS active devices at mm-wave frequencies. Even though
theCurtice-Ettenberg model is traditionally used to model MESFETs, it was found to be a useful tool in accurately predicting the performance trends in the Doherty. The finite output impedance was found to have the biggest impact on the performance and it can reduce the ideal Doherty PAE by as much as 25 %. Finally, a 71-76 GHz DPA is proposed and implemented in a 90-nm RF-CMOS process. The footprint of the DPA is 1.53 mm2 and it has a P1dB of 11.7 dBm, a PAE of 30.6 % and a PAE at 6-dB back-off of 15.6 %. At P1dB the DPA is consuming 21.7 mA from the 1.5 V supply. The reconfigurable option improves the gain by 3.2 dB and draws an additional 7.1 mA. This option makes the circuit more functional at mm-wave frequencies. The efficiency characteristics of the implementation are within 5 % of the model predictions.
ACKNOWLEDGMENT
The authors would like to thank the Communications Research Center Canada (CRC) for providing the design facilities, NSERC for the funding, and the Canadian Microelectronics Corporation (CMC) for the