A 335–407-GHz SiGe-Based Subharmonic Mixer Using a Fully Integrated LO Generation

This letter introduces a silicon–germanium (SiGe)-based subharmonic mixer (SHM) optimized for THz radar applications, operating at a center frequency of 360 GHz with a 3-dB bandwidth of 72 GHz. This mixer is a key component in a fully integrated receiver, featuring a wideband 90-GHz voltage-controlled oscillator (VCO), a frequency doubler stage, and power amplifier (PA) stages. The mixer exhibits a conversion gain of −6.1 dB while maintaining an excellent input compression point of >4 dBm, accompanied by a simulated noise figure (NF) of 24.8 dB. Moreover, a current consumption of only 8.6 mA underlines the energy efficiency.

Furthermore, the detection of tiny particles [6], [7] and the classification of materials, either solid [8] or gaseous [9], is enhanced by the use of THz signals.Finally, THz radiation offers significant advantages in medical [10], [11] and security [12], [13] applications.In THz research, silicon technologies are vital, since they enable highly integrated solutions by combining the digital and RF domains, including on-chip antennas, into one chip.Moreover, integrated on-chip antennas can eliminate off-chip transitions and the need for RF PCBs [14] and show improved efficiency when increasing the frequency to THz [15], [16].
This letter introduces a 360-GHz subharmonic mixer (SHM) with an on-chip local oscillator (LO) generation for FMCW radar applications using the 90-nm B12HFC silicon-germanium (SiGe) BiCMOS technology from Infineon [2], [17], featuring an f T / f max = 300/500 GHz.The circuit design and fabricated MMIC are detailed in Section II.Section III analyzes the mixer's simulation and measurement results, which are compared against other SiGe mixers in Section IV.Finally, Section V concludes this letter.

II. CIRCUIT DESIGN
THz transceiver circuit designs often employ harmonic approaches due to the technology's transit frequency limitations.While SHM circuits may experience reduced conversion efficiency [18], they eliminate the need for additional frequency doubler or PA stages [19].This trade-off can result in a more compact MMIC footprint and lower power consumption.
Fig. 1 shows the block diagram and the breakout chip, including a transmit (Tx) path and a receive (Rx) path.While the 360-GHz Tx path was presented in [20] and [21], this letter describes the Rx path depicted in the block diagram and focuses on the SHM.
A Colpitts-Clapp VCO on the breakout MMIC generates a 90-GHz LO signal.This signal is subsequently doubled to 180 GHz using a Gilbert-cell-based frequency doubler.The 180-GHz signal is then amplified by a two-stage PA, as described in [20] and [21].After amplification, the LO signal is split and routed to both the Tx path, described in [21], and Rx path [see Fig. 1(b)].The Rx path begins with a 90 • hybrid coupler, generating in-phase (I ) and quadrature-phase (Q) signals (detailed in [21]).These I Q signals are then amplified by the same two-stage PAs and fed into the SHM, whose size is 200 × 140 µm 2 .To characterize the SHM,  a synthetic 360-GHz Rx signal is applied using single-ended RF pads and an on-wafer probe.This signal is then converted to a differential signal using a rat-race balun.
The circuit diagram of the balanced SHM is shown in Fig. 2, highlighting key sections: the mixer core, the LO bias network for the HBT pairs, the current mirror circuitry, and the matching networks for the I and Q inputs.The mixer core utilizes a Gilbert-cell topology with an integrated push-push- based frequency doubling mechanism, enabling subharmonic operation.
To maximize conversion gain, several design choices were implemented in conjunction.First, the RF signal is directly coupled to the mixer core, bypassing HBTs (typically used as transconductance stages).Since the HBTs operate above their f T , they cannot provide current gain.In addition, direct coupling can improve broadband matching [19].
Omitting the transconductance stages increases the available power supply headroom for utilizing large load resistors to enhance conversion gain.However, a limited 3.3-V power supply and large resistors could lead to suboptimal current densities (impacting f T ) and inadequate base-collector voltages (affecting compression behavior).We found that 1200polysilicon resistors provide a good balance, meeting our design criteria.
Within the mixer core, we utilize compact 1.8-µm BEBC HBTs.These allow operation with low dc currents while maintaining sufficient f T and facilitating the use of large load resistors for improved conversion gain.Finally, we jointly designed the matching network of the LO and Rx port to maximize the conversion gain, as their matching influences each other.
Inside the layout of the mixer, we had to cross-multiple transmission lines (TRLs) caused by the Gilbert-cell topology itself and by integrating a push-push topology using I Q signals.The LO and Rx matching networks employ the highest copper layer (M7) and the overlying aluminum layer for TRLs.The connections among the eight collector nodes are established beneath the Rx TRLs using M6-based embedded striplines, with M2 and M7 as the ground plane.

III. SIMULATION AND MEASUREMENT RESULTS
Fig. 3 depicts the balun's transmission behavior, including amplitude and phase difference and insertion loss.The input matching is not shown but is better than −17 dB for every port.A modified Lange coupler creates I Q signals, utilizing a vertical TRL coupling topology to minimize its physical size.The simulation shows a phase error of 4 • -6 • and an amplitude imbalance just below 3 dB [21].
The circuit simulation considers RC parasitics for the resistive load, current mirror, and mixer core.Moreover, it accounts for self-heating using 80 • C device temperature and uses EM simulation results for all TRLs, which are not covered by the manufacturer's models.Due to stability issues in the harmonic Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.balance simulation, we excluded the PAs and couplers and had to use ideal I Q ports.
In Fig. 4, the simulated CG is plotted for varying LO and Rx power levels at a constant IF frequency of 100 MHz.Increasing LO power improves SHM performance until it saturates and then declines from ≈3 dBm onward.Also, excessive Rx power degrades the CG.The 1-dB input compression points were determined at ≈9 dBm (310 GHz), ≈5 dBm (360 GHz), and ≈4 dBm (410 GHz).As frequency rises, compression behavior worsens due to improved Rx matching.Fig. 5 shows the simulated CG and noise figure (NF) for an LO power of 0, 5, and 10 dBm.The simulations show a CG of up to 3.4 dB and a double-sided NF of 24.8 dB when using ideal I Q ports.
We measure the conversion gain using a synthetic receive signal generated by a WR2.2 VNA extender from VDI and a Keysight analog signal generator (E8257D).The IF signal is measured with an oscilloscope (Keysight MSOS804A), which allows for measuring the receiver with its unstabilized VCO in time domain.
In Table I, we use 3-dB bandwidth, CG, double-sided NF, and power consumption as comparison criteria.Note that conversion gain in some publications includes an IF buffer or an IF amplifier; we deliberately omit this to use SMD amplifiers on the PCB in our upcoming system.Our design demonstrates impressive bandwidth and power consumption characteristics.Notably, it achieves competitive conversion gain despite the absence of an IF buffer or amplifier.We optimized the LO power for peak gain, a capability not available in all compared mixers, contributing significantly to our design's performance.This optimization, the advanced B12HFC SiGe technology, and other design choices contribute to the achieved values of the presented mixer.

V. CONCLUSION
The presented SiGe-based SHM, designed for THz radar applications, offers a competitive conversion gain of −6.1 dB without additional amplification, an impressive input compression point exceeding 4 dBm, and low-power consumption of 8.6 mA at 3.3 V.These attributes make it well suited for THz radar sensors, which are intended for our future research.

Fig. 2 .
Fig. 2. Circuit of the SHM, including the matching network, the current mirror, and the bias network.External signals are marked in red, and internal signals are marked in black.The current density of the HBTs can be adjusted by applying a voltage to V I 0 .

Fig. 3 .
Fig. 3. Simulation results of the rat-race balun showing the amplitude and phase difference, insertion loss, and transmission behavior using sonnet.The circular marker shows the phase difference (right axis).

Fig. 5 .
Fig. 5. Simulated NF and conversion gain (RC parasitics, 80 • C device temperature, ideal I Q, and Rx signals) and measured conversion gain of the SHM (synthetic Rx signal using a WR2.2 frequency extender and f IF ≈ 100 MHz).

TABLE I SIGE
-BASED MIXERS BETWEEN 300 AND 500 GHz we determined the LO to Rx isolation using an H -band spectrum analyzer and probe contacting the Rx pads and measured less than −40 dBm.