Design of a 240 GHz on-chip dual-polarization receiver for SIS mixer arrays

Our receiver chip is fully planar, formed mainly from microstrip transmission lines. The common microstrip topology eases the design and connection of on-chip circuit elements without transitions to other transmission line topologies while providing a common ground throughout the chip. The circuit is fabricated on a quartz substrate with a relative permittivity $\epsilon_\mathrm{r} = 3.78$ in a 5-step process, summarized in table 1. In the first step, the 14 kA cm⁻² Nb/AlO_x/Nb trilayer is deposited on all areas covered with the ground conductor. In the second step, the trilayer is etched to the 1.5 µm² SIS junctions, leaving the bottom layer on the remaining areas to form the 400 nm thick Nb ground-plane layer. After etching, a first SiO layer is deposited in the same fabrication step so that the SIS junctions are self-aligned with the SiO layer and the entire ground-plane layer is covered. In the third step, a second SiO layer is deposited, forming a 200 nm deep 6 µm by 6 µm well where the SIS junctions are exposed to the 400 nm thick wiring layer, which is deposited in the fourth step. In the final step, the bonding pads are coated with 150 nm Au for reliable wire bonding to the chip. After completion of the deposition, the quartz substrate is diced and thinned to 50 µm. Moreover, except for the SIS junctions, all circuit features are designed to be larger than 3 µm to ensure a reliable photolithography process. This, together with the 400 nm thick SiO ($\epsilon_\mathrm{r} = 5.7$) layer, defines a reliably fabricated characteristic impedance of 19 Ω for the microstrips. The superconductivity of Nb at 4.2 K is represented by a frequency dependent surface impedance of a Bardeen Cooper Schrieffer (BCS) superconductor with a Cooper pair binding energy equivalent voltage $V_\mathrm{gap} = \frac{2\Delta}{e} = 2.8\,\mathrm{mV}$ and normal state resistivity $\rho_\mathrm{N} = 8.13$μΩ cm [20].

We use a planar OMT to couple the RF and LO signals into microstrips on the receiver chip while separating the two polarizations, similar to bolometric detectors of cosmic microwave background experiments [21–25]. For this setup, the free-space signals couple via a feedhorn into two perpendicular TE₁₁ modes in the ⌀$1.0\,\mathrm{mm}$ circular waveguide. These two modes are then guided onto the planar circuit with two sets of opposing probes forming the OMT shown in figure 4. We employ a probe set per polarization so that each probe receives half of the incident signal power of one polarization with 180^∘ phase difference between the opposing probes due to the symmetry of the TE₁₁ field distribution.

Zoom In Zoom Out Reset image size

The OMT probes are designed to couple maximum power from the circular waveguide with an impedance of a couple of hundred Ohms to the 19 Ω microstrips. The backshort extension of the circular waveguide further enhances the coupling into the probes and, thus, the microstrips.

The presence of the quartz substrate causes reflections of the signal at the interfaces. Although thinning the substrate to 50 µm (which is the practical limit considering the brittleness of quartz) mitigates these effects, the losses are still noticeable. These losses could potentially be minimized using silicon-on-insulator (SoI) technology, but this is still under development in our facilities.

In addition to losses through the substrate, losses also occur in the vacuum slit above the substrate. The slit needs to exceed a minimum height to avoid interference with the on-chip circuit and to accommodate fabrication tolerances, such as the substrate thickness and the machining of the pixel. Balancing the OMT performance with the losses and the uncertainties in the fabrication results in a 50 µm slit.

The losses through this slit are reduced by a quarter-wavelength deep choke groove, machined concentrically at a quarter-wavelength distance from the circular waveguide. This causes the electromagnetic wave in the circular waveguide to encounter an electric short, which prevents a significant fraction of signal power from leaking into the vacuum slit.

Both of the losses discussed above propagate as independent parallel-plate waveguide TEM modes in the substrate and the vacuum slit since these two volumes are separated by the ground-plane layer of the microstrip. The choke groove forms a vacuum-filled parallel-plate ring between the circular waveguide and the choke groove. This ring resonates when an integer number of wavelengths fit into the ring circumference, preventing those signals from coupling into the OMT probes. Adjusting the radius of the circular waveguide and the choke groove changes the ring circumference and therefore offers a method of shifting the resonance to frequencies outside the receiver bandwidth. However, as the resonance is shifted away from the operation band, another order resonance will shift towards the operation band from the other frequency end. Thus, the choice of waveguide and groove radii prevents resonances up to a certain fractional bandwidth. The wSMA specifications of 42% exceed this fractional bandwidth.

To circumvent this problem, Mauskopf et al [26] increased the circumference length of the parallel-plate ring by cutting radial slots between the circular waveguide and the choke groove at a 45^∘ angle to the OMT probes, removing the resonance from the frequency band. However, this solution complicates the machining process of the pixel. We therefore disturb the parallel-plate waveguide with radial serrations in the ground-plane layer, shifting the resonance to a frequency outside the band. In summary, we limit the losses through the vacuum slit with a choke groove and suppress the emerging resonances by serrating the ground-plane layer.

Figure 5 shows the simulated response of the OMT. The coupling to the individual probes is essentially identical, leading to a uniform coupling for both polarizations on the chip. This is crucial in order to avoid receiver biases towards one polarization. In addition, the cross-coupling generated by the OMT is less than $-30\,\mathrm{dB}$. The return loss is reasonable at less than $-10\,\mathrm{dB}$, and the total coupling of each polarization into the designated microstrip pair is better than $-2\,\mathrm{dB}$. This not-optimal coupling is a consequence of the unavoidable $-6\,\mathrm{dB}$ loss due to the relatively thick substrate, which can be improved with SoI technology in the near future.

Zoom In Zoom Out Reset image size

Routing the full signal power of each polarization from the OMT to the hybrids inevitably leads to a crossing of two transmission lines carrying different polarizations due to the OMT geometry, which couples the polarizations into the microstrips connected to opposing probes. Therefore, the crossover must be designed to minimize crosstalk between the two crossing transmission lines.

Our planar broadside coupler solution, shown as an inset in figure 6, is largely similar to the design reported in [25, 27] but now at a much higher centre frequency and with lower characteristic impedances. The microstrips are transformed into two coplanar waveguide (CPW) structures in the two conductor layers of the microstrip, namely the wiring layer and the ground-plane layer. In the wiring layer, the CPW is formed from two conductor patches bracketing the 3 µm wide microstrip feed with 3 µm wide gaps over a length of 125 µm, and the ground-plane layer has an opening to deposit the 4.4 µm wide centre strip between the 3 µm wide gaps of the 69 µm long ground-plane CPW. This CPW connects to the feeding microstrip via two 25 µm by 15 µm broadside coupler sets, DC-isolated conductor rectangles in the ground-plane layer and the wiring layer, complying with our 5-step fabrication process. In addition to carefully selecting the broadside coupler dimensions, the performance is enhanced by providing an additional load to the broadside couplers in the wiring layer with the patches bracketing the wiring-layer CPW.

Zoom In Zoom Out Reset image size

The central conductors for the CPWs overlap less than 15 µm², leading to the excellent cross-coupling isolation shown in figure 6. This low cross-coupling and return loss yield a high transmission using this compact and easy-to-fabricate crossover design for routing the OMT outputs to the hybrids for recombination.

Once the two microstrips from an OMT probe set are routed to the same chip location, a circuit element is required to recombine those to feed the mixer with the maximum power available. We abstained from employing a 3-port circuit solution due to the difficulty of implementing power-dissipating circuit elements with our superconducting thin-film deposition process and fundamental impedance mismatches arising from non-dissipative 3-port circuits. Instead, the two-section 90^∘ branch-line hybrid shown in figure 7 is implemented to recombine the signal since all four ports of this unidirectional circuit can be matched simultaneously [28–30].

Zoom In Zoom Out Reset image size

The hybrid branches are approximately quarter-wavelength long microstrips with carefully selected characteristic impedances, with the narrowest 19 Ω microstrips connecting adjacent ports, 5 Ω microstrips leading from the ports to the centre and a 6 Ω microstrip in the centre between the two hybrid sections. The branches are arranged symmetrically in two squares, where the additional section widens the frequency bandwidth, as shown in figure 7.

An input signal at the Σ port couples at $-3\,\mathrm{dB}$ to the Through and Coupled port with a 90^∘ phase difference while leaving the Δ port isolated. As the branch-line hybrid is unidirectional, two identical signals with 90^∘ phase difference applied at the Through and Coupled port consequently superimpose on the Σ port constructively while superimposing deconstructively on the Δ port. Hence, the hybrid is suitable for recombining the signals splitting between an OMT probe set.

Each hybrid output connects to the SIS mixer via a BPF, preventing IF signals from leaking into the RF circuit while maximizing the IF power output. Furthermore, the BPF isolates the DC bias of the individual SIS mixers while passing the RF signal from the hybrid output to the SIS mixer. In this way, we can DC bias each SIS mixer independently.

The BPF design, shown in figure 8, utilizes broadside couplers to route the signal from the 3 µm wide microstrip into a 80 µm long CPW deposited within an opening in the ground-plane layer leaving 5 µm wide gaps, similar to the crossover. Again, a 72 µm by 25 µm conductor patch on top of the CPW is deposited to improve the coupling of the broadside couplers. Consequently, the RF couples efficiently from the hybrid to the following SIS mixer with close to 0 dB coupling and without DC and IF leakage from the SIS mixer into the hybrid.

Zoom In Zoom Out Reset image size

SIS junctions suffer from a high geometric capacitance, and this capacitive impedance needs to be matched to the embedding circuit. Although a single junction's impedance can be tuned with a series of stubs, we opted for a twin-junction tuning circuit in which two SIS junctions are connected in parallel with a microstrip section [31], as shown in figure 9. Additionally, a 139 µm long and 9 µm wide series microstrip transformer is added in front of the twin-junction circuit to improve the impedance matching to the circuit prior to this SIS mixer circuit over the full bandwidth.

Zoom In Zoom Out Reset image size

The downconverted IF signal is routed to the bonding pad via a microstrip connected to the second SIS junction. This microstrip would also pass considerable RF power to the bonding pads, reducing the power available for downconversion at the SIS junctions. Therefore, the second SIS junction is connected directly to a LPF, which consists of alternating high- and low-impedance, 3 µm and more than 50 µm wide microstrip sections to choke RF signals while passing IF signals. This circuit, feeding the IF to the bonding pad, has an associated input impedance, which alters the RF tuning of the twin-junction circuit. Hence, the twin-junction tuning design must account for the LPF circuit.

The tuning circuit shown in figure 9 comprises two circular 1.5 µm² SIS junctions with 13.3 Ω dynamic resistance and 120 fF geometric capacitance separated by a 46 µm long 18 Ω microstrip section. The LPF reflects the RF signal power so that more than $-0.4\,\mathrm{dB}$ of the incoming RF power couples into the two SIS junctions and less than $-30\,\mathrm{dB}$ couples through the LPF. Both SIS junctions couple close to $-3\,\mathrm{dB}$ of the input power at the centre frequency, while on each end of the band, the signal couples to one of the junctions predominantly.

The polarized signal coupling network, shown in figure 10, contains the OMT, crossover, and two hybrids. The crucial parameter in this circuit is the microstrip length between the OMT probes and the hybrid input ports. To ensure that the full signal power of one polarization recombines at the Σ port of the hybrid, the microstrip lengths adjust the 180^∘ phase difference between a set of opposing OMT probes to the 90^∘ phase difference at the inputs of a hybrid by having a length difference equivalent to a quarter-wavelength at the centre frequency.

Moreover, the routing of the two polarizations is symmetric to avoid undesired biases towards one polarization due to loss differences introduced by, for example, the dielectric. Therefore, the crossover is placed at the centre between OMT probes, and the hybrids are located symmetrically with respect to the centre line formed by the OMT and crossover. Furthermore, the hybrids are arranged in parallel to this centre line so that their outputs lead to the same chip edge at which all mixers can be connected to the IF transformer, keeping the receiver compact.

We designed the transmission lines connecting the OMT to the hybrid with straight and curved microstrip lines to avoid corner discontinuities and to reduce the total microstrip lengths. To make the optimization of the design feasible, we first chose a suitable geometry with which all microstrips could be restricted except the straight sections connecting to OMT probes and the following arc, which leads into a concentric circle centred at the OMT, as shown in figure 10. We then fine-tuned the 90^∘ phase difference by varying the length of the straight microstrip sections connecting the OMT probes and the radius of the following curved section, keeping the radius of the OMT-concentric circle, the position of the crossover, and hybrid fixed.

Table 2 summarizes the physical lengths of the microstrips connecting the OMT probes and the hybrids in figure 10, which are routed at the crossover in the ground-plane-layer CPW for Pol. 1 and the wiring-layer CPW for Pol. 2. The length of the microstrips routed via the crossover is approximately a quarter-wavelength shorter than that of the uncrossed microstrips; thus, the Σ ports of the hybrids are closer to the OMT-crossover centre line.

In addition to adjusting the phase difference at the hybrid inputs, all circuit elements are impedance matched to avoid reflections, maximizing the signal power coupling to the mixer. Both the OMT and the crossover connect to 19 Ω microstrips, requiring no intermediate impedance transformer, whereas 9 Ω microstrips connect the hybrid. Consequently, the microstrips connecting the OMT and crossover to the hybrids are designed as 5-section binominal impedance transformers [29]. Similarly, the hybrid outputs are constructed from 3-section binominal impedance transformers to match the BPF in the mixer circuit, which again is fed by a 19 Ω microstrip.

The simulation in figure 11 shows that both polarizations couple almost identically with approximately $-2\,\mathrm{dB}$ to their designated Σ ports. In fact, the cross-coupling between the two Σ ports, corresponding to contamination of the two polarizations, is less than $-30\,\mathrm{dB}$. Furthermore, the low coupling level into the Δ ports suggests that the circuit is indeed tuned to a 90^∘ phase difference at the hybrid inputs. Similarly, the general behaviour of the OMT remains unchanged apart from the expected addition of return loss poles and a slightly higher return loss level, which can be attributed to the addition of impedance transformers and circuit elements such as the hybrid. In summary, the polarized signal coupling network reliably separates the polarizations and recombines the signal power into a single microstrip throughout the desired frequency band.

Zoom In Zoom Out Reset image size

The polarized signal coupling network outputs connect to the mixer circuit, comprising the BPF, the SIS mixer and the LPF. We employ these filters so that the RF signals can propagate via the BPF to the SIS mixer for downconversion but not through the LPF, and the downconverted IF signal can only propagate through the LPF.

Figure 12 shows that the BPF connects to the RF input of the SIS mixer via a 3-section impedance transformer to match the 19 Ω output of the BPF to the 12.5 Ω input of the SIS mixer circuit. The impedance transformer and the LPF are meandered to decrease the footprint of the mixer circuit, while the twin-junction circuit of the SIS mixer is on a straight microstrip section. This layout can be fitted to the polarized signal coupling network while maintaining compactness and symmetry. Likewise, the LPF output is angled compared to the feeding microstrip to achieve even spacing between the bonding pads in the layout of the whole receiver chip.

Zoom In Zoom Out Reset image size

Figure 13 shows the power coupling from the junctions to the IF and RF ports as a function of frequency. Adding the BPF to the SIS mixer and LPF introduces poles in the return loss, but none of these poles peak above $-10\,\mathrm{dB}$ within the operation band. The main characteristics of the coupling into the individual junctions remain unchanged at the RF range. In fact, the addition of the BPF reduces the total coupling by less than $-0.2\,\mathrm{dB}$ while introducing a $-30\,\mathrm{dB}$ isolation between the RF and IF port throughout the shown IF and RF range. At low IF, each junction couples $-3\,\mathrm{dB}$ to the IF output, and the total coupling summed over both junctions only drops below $-3\,\mathrm{dB}$ above 30 GHz, suggesting the potential broad IF bandwidth of our receiver.

Zoom In Zoom Out Reset image size

The performance of the entire receiver chip, shown in figure 14, is simulated by connecting the polarized signal coupling network to four mixer circuits in Ansys circuit. The IF response is identical for all four mixer circuits, and the polarized signal coupling network does not affect the IF behaviour, indicating a well-functioning BPF. At RF, both polarizations continue to behave similarly, without any bias towards a particular polarization. In fact, adding the mixer circuits introduces only extra return loss poles without overly increasing the return loss level and increases the coupling into the Δ mixer circuit by less than 2 dB. The coupling level from the circular waveguide to the designated Σ mixers is around $-2\,\mathrm{dB}$, dominated by the OMT losses, as explained earlier. The cross-coupling level between the two polarizations remains below $-30\,\mathrm{dB}$, mirroring the behaviour of the polarized signal coupling network.

Zoom In Zoom Out Reset image size

We analyse the mixer gain and noise temperature of our receiver using a circuit model similar to the Ansys circuit setup, with a C++ script based on the SuperMix library [33]. The circular waveguide has a frequency-dependent port impedance following HFSS simulations, but now we also employ a standard impedance matching circuit to match the on-chip circuit to 50 Ω. We chose the LO power and DC bias in order to maximize the double-sideband (DSB) mixer gain at 5 GHz IF. Figure 15 shows the optimized DSB mixer gain, the sum of lower-sideband and upper-sideband gain, and preliminary predictions for the noise temperature. Again, both polarization states have an almost identical response. The DSB gain approaches 0 dB, and the DSB noise temperature is between two and three times the quantum limit $T_\mathrm{Q}$. We want to emphasize that the noise temperature values shown in figure 15 are much lower than expected in the real experiment since they do not include losses of several receiver components such as optics; therefore, they only need to be used as a guide.

Zoom In Zoom Out Reset image size