Fabrication-free assessment of microwave losses in germanium-based dielectrics and superconductors

AI summary

Overview

• Research area: Quantum physics / superconducting quantum circuits and semiconductor–superconductor hybrid devices, specifically microwave loss characterization of materials for germanium-based quantum technologies.
• Technical level: Intermediate (readers should be comfortable with resonators, quality factors, and loss tangents; the materials science is explained in the text).
• Scope: The paper introduces a flip-chip sensing scheme that measures microwave losses of dielectric and superconducting target materials without fabricating any device on them, and applies it to Ge/SiGe quantum-well heterostructures, PtSiGe, and Nb-coated PtSiGe.

What This Paper Is About

Fully fabricated microwave resonators on planar Ge heterostructures typically show high losses, but the microscopic origin of those losses is unclear because substrate, deposited dielectrics, superconductor choice, and interfaces all contribute and are hard to separate. The standard way to disentangle them is to build many resonator geometries from the material of interest, which is slow and itself introduces processing damage. The authors instead place an unpatterned piece of the material on top of a superconducting sensor chip and read out how much extra loss and kinetic inductance it adds, giving a fast, fabrication-minimal screen.

Key Contributions

1. A fabrication-free microwave loss sensor. A flip-chip scheme in which a sensor chip with eight λ/4 coplanar waveguide resonators (100 nm Ta on a 5 nm Nb seed layer on Si; center metal width 6 µm, gap 12 µm; bare resonance frequencies from 4.2 GHz to 7.2 GHz) is pressed against an unpatterned target chip using 4 µm-tall, 100 µm-radius SU-8 pillars and pogo pins, with no indium bumps and no device fabrication on the target beyond dicing or cleaving.
2. Quantified dielectric losses of Ge-based materials using the extracted material participation ratio, including Ge substrates, a strain-relaxed Ge₀.₇₅Si₀.₂₅ virtual layer (QW etched), an as-grown Ge/SiGe quantum well heterostructure, and HF-treated versions of both, with cross-validation by an on-chip Nb coplanar strip resonator.
3. Microwave characterization of the superconductor PtSiGe formed by thermal reaction of deposited Pt with the heterostructure, together with a Nb-coated variant (Nb-PtSiGe) formed from an Nb/Pt bilayer, including STEM-EELS structural analysis, transport critical temperatures, microwave loss, aging, and kinetic inductance.
4. Design guidance and a screening platform: the results constrain which materials and surface treatments are compatible with low-loss microwave circuitry on planar Ge, and the method is presented as a general fast-turnaround test for new materials.

Main Findings

• Sensitivity is tunable via pillar height. For Ge (ε_r ≈ 16.2) the dielectric participation ratio can be tuned from 6.29% to 0.62% by changing SU-8 pillar height from 2 µm to 10 µm; with the 4 µm pillars used, p_r,material = 2.62%, corresponding to a detectable loss tangent range of 20×10⁻⁶ to 1000×10⁻⁶.
• The sensor probes the near-surface region but also sees bulk. About 26% of the electric energy in the target dielectric is stored within its top 2 µm.
• Bare sensor baseline. The intrinsic loss of the sensing resonator with no target ("vacuum") is 1.5×10⁻⁶.
• Ge and the SiGe virtual layer are similar. Extracted low-photon material loss tangents are (199 ± 40)×10⁻⁶ for the Ge substrate and (197 ± 29)×10⁻⁶ for the heterostructure etched down to the Ge₀.₇₅Si₀.₂₅ virtual layer. The authors note Ge losses are substantially higher than the 0.01×10⁻⁶ reported for bulk float-zone Si.
• On-chip measurement confirms the flip-chip result. A Nb coplanar strip resonator (15 µm gap, 10 µm center trace) on the same etched heterostructure gave a total low-power loss of (208 ± 19)×10⁻⁶; with a simulated substrate participation of 93.9%, the extracted material loss tangent is (222 ± 20)×10⁻⁶, 10% higher than the flip-chip value.
• As-grown quantum wells lose more, and HF helps. The as-grown QW heterostructure gave δ_material^LP = (678 ± 457)×10⁻⁶, roughly three times higher than pure Ge and the QW-etched sample, and it showed finite residual conductance at cryogenic temperatures. HF treatment reduced this to (102 ± 54)×10⁻⁶, comparable to HF-treated Ge at (82 ± 51)×10⁻⁶; HF reduces δ_LP by about a factor of 2 and δ_HP by an even larger factor.
• Power dependence differs. Ge shows a relatively flat power dependence compared with the QW-etched sample, which does not saturate at ⟨n⟩ ≈ 1; the authors attribute this to a lower intrinsic two-level-system concentration in the commercial Ge substrate versus grading, interfaces, and/or etching.
• PtSiGe structure. STEM shows a 138 ± 7 nm PtSiGe layer with no distinct Ge QW remaining, indicating the reaction reaches the buried QW (66 nm deep in this heterostructure); surface roughness is 4.3 nm.
• Nb coating changes structure and transport. Nb-PtSiGe consists of a 37 ± 3 nm Nb layer on 128 ± 11 nm PtSiGe, with roughness reduced to 2.4 nm; Ar bubbles are observed between the layers, likely from sputtering.
• Critical temperatures. Transport gives T_C^DC = 0.54 K for PtSiGe and 1.5 K for Nb-PtSiGe — a nearly three-fold enhancement.
• Reference superconductor losses. A 60 nm Nb film on Si gave sensor losses δ_LP = (1.5 ± 1.0)×10⁻⁶ and δ_HP = (0.5 ± 0.3)×10⁻⁶; 60 nm Nb on bare Ge gave δ_LP = (6.1 ± 0.6)×10⁻⁶ and δ_HP = (3.9 ± 0.7)×10⁻⁶.
• PtSiGe has high and unstable microwave loss. As-grown PtSiGe: δ_LP = (158 ± 141)×10⁻⁶, rising to (736 ± 112)×10⁻⁶ after two months of ambient exposure; a buffered oxide etch (BOE) treatment on a fresh film, confirmed by XPS to remove some surface oxide, gave (65.3 ± 9.7)×10⁻⁶.
• Nb coating strongly suppresses loss and aging. Nb-PtSiGe: δ_LP = (13.2 ± 2.0)×10⁻⁶, rising only to (18.1 ± 1.7)×10⁻⁶ after two months.
• RF critical temperatures are below transport values. Mattis–Bardeen fits to the high-power loss versus temperature give T_C^RF = 0.30 K for PtSiGe and 0.33 K for Nb-PtSiGe (assuming Δ₀ = 1.764 k_B T_C), with fit parameters A_QP = 0.1, 1/Q_other = 550×10⁻⁶, f_r = 7.3 GHz for PtSiGe and A_QP = 0.1, 1/Q_other = 14.3×10⁻⁶, f_r = 4.7 GHz for Nb-PtSiGe. The disagreement with transport T_C is consistent with gap inhomogeneity in both films.
• Nb-PtSiGe is not yet resonator-ready on its own. A fabricated Nb-PtSiGe coplanar strip resonator, designed for a coupling quality factor of about 3000, produced no measurable resonance, implying an internal quality factor more than 10× lower; the authors estimate sidewall metal-air participation around 46×10⁻⁶ and an internal quality factor of around 500 from oxide loss alone, leaving the metal-substrate interface as a likely additional channel that the flip-chip method cannot probe because electric fields are screened.
• Kinetic inductance is measurable. The sensor extracts L_K□ over 0–10 pH/□ with roughly a 1% resonance-frequency shift per 1 pH/□; a 50 nm δ-NbN calibration film gave 6.9 ± 1.1 pH/□ versus 7.7 pH/□ calculated from transport, PtSiGe gave 3.8 ± 0.2 pH/□, and Nb-PtSiGe gave 3.0 ± 0.2 pH/□.

Methodology in Plain English

The team built a small chip holding eight superconducting microwave resonators whose electromagnetic fields fringe out of the chip surface. They patterned SU-8 photoresist pillars (4 µm tall, 100 µm radius) as standoffs, then flipped the material to be tested on top and pressed it down with pogo pins inside the package; the pillars hold a fixed, repeatable gap (pillar height changed by less than 3% after 10 thermal cycles). Measurements were done in a dilution refrigerator at a mixing chamber temperature of 10 mK unless stated otherwise.

Because the resonator's electric field reaches into the target, any loss in the target adds to the resonator's measured loss. The authors subtract the bare sensor loss, then divide by the simulated fraction of electric energy stored in the target (the participation ratio) to get a material-level loss number. The same field-reaching geometry makes the target's kinetic inductance shift the resonance frequency, which provides a second, complementary readout. For superconductors, the relevant extra contributions are conductor loss and loss in the native oxide at the metal-air surface, the latter simulated with a 1 nm oxide of ε_r = 10 giving p_r,MA = 4.3×10⁻⁶, comparable to the sensor's own metal-air participation of 6.8×10⁻⁶. Control experiments included a bare "vacuum" measurement, a conventional Nb film for calibration, and a fully fabricated on-chip resonator on the same heterostructure to check the flip-chip numbers. Structural and chemical characterization used STEM-EELS and XPS, and transport measurements gave DC critical temperatures.

Why This Matters

The work separates loss sources that are normally entangled in fully processed devices, and it does so without exposing the material under study to resist, etchants, solvents, or handling — factors the paper notes can themselves degrade microwave loss. It establishes concrete constraints: the strain-engineered Ge/SiGe platform with a virtual layer grown directly on Ge does not show substantially worse loss than the Ge substrate, surface chemistry (residual conduction, suboxides) can dominate, and PtSiGe — despite forming highly transparent contacts to Ge quantum wells — is a poor standalone superconductor for low-loss resonators unless overcoated.

Real-world applications:

• Screening candidate gate dielectrics, substrates, and superconducting films for semiconductor-spin-qubit and hybrid superconductor–semiconductor quantum processors before committing to a full fabrication run.
• Guiding microwave circuit and package design on planar Ge heterostructures, including which interfaces and surface treatments to prioritize.
• Rapid quality control of thin films and heterostructures in materials-growth and foundry settings, using only a diced chip.
• Providing kinetic inductance values (pH/□) needed for designing impedance, nonlinearity, and coupling in superconducting circuits.

Industry relevance: semiconductor quantum computing is moving toward 300 mm-compatible process lines, where each fabrication loop is expensive; a dicing-only measurement that returns a loss tangent and kinetic inductance in a short cycle directly addresses materials-selection bottlenecks for both quantum hardware and cryogenic electronics more broadly.

Future Directions

• Varying the resonator cross-sectional geometry or the resonator–target distance to separate bulk, surface, and interface loss channels more directly.
• Applying the same sensor to other Ge/SiGe heterostructure designs — including those avoiding a plastically relaxed Ge starting layer, motivated by the 4.7% Si/Ge lattice mismatch — so that platforms can be compared without differences in fabrication.
• Identifying surface preparations beyond HF that suppress residual conduction and suboxides across heterostructures grown by different groups.
• Exploring alternative low-loss superconductors: other reaction-based metal–SiGe films, planar-tunneling-based proximity methods, or other bilayers, plus process changes such as Kr-based sputtering to reduce Ar bubbles; the unprobed metal-substrate interface loss in Nb-PtSiGe resonators is flagged as a target for future study.

Target Audience

Materials scientists and condensed-matter physicists working on superconducting qubits, spin qubits, and hybrid superconductor–semiconductor devices; microwave engineers designing resonators and flip-chip packages; and process engineers who need a fast, low-fabrication method to screen new dielectrics and superconducting thin films for cryogenic and quantum applications.

Authors’ abstract

We present a flip-chip-based sensing scheme to measure effective microwave losses associated with target materials for quantum technologies, without requiring any device fabrication on the material under test. Using this approach, we quantify the microwave losses of a strain-engineered Ge/SiGe quantum well heterostructure and investigate losses arising from its Ge substrate and intermediate layers. The quality factors of the fabricated microwave resonators agree with the losses of dielectric materials independently extracted from flip-chip sensing measurements. We further study the superconductor platinum silicon germanide (PtSiGe) prepared by thermal reaction with a deposited Pt film, finding high microwave losses that limit the suitability of the films studied here as the sole superconductor for high-quality resonator applications. By coating Pt with Nb prior to the reaction, we observe a substantial reduction in microwave loss and a nearly three-fold enhancement of the transport critical temperature. The temperature dependence of the microwave loss is consistent with gap inhomogeneity in both superconducting films. These results identify constraints on material choices, provide design guidance for microwave circuits on planar Ge heterostructures, and demonstrate a fast-turnaround testing method for new materials for superconducting quantum circuits.