Introducing: A Motherboard for Quantum Computers

By Eric Michiels

As quantum systems scale up, high fidelity and high-speed qubit readout—without bulky components—is a critical success factor. An IBM Research Team led by Baleegh Abdo has demonstrated the proof-of-principle of a high-performance qubit readout motherboard, free from the disturbances of hardware like circulators and isolators.

Reading out qubits is a huge challenge: stable superconducting qubits operate in a dilution refrigerator at 15 mK, but must be reliably measured using room-temperature electronics.  In order to carry out measurement today, quantum computers employ low-power microwave signals and then amplify those outgoing signals using Josephson Parametric Amplifiers (JPAs) followed by High Electron Mobility Transistor (HEMT) amplifiers—learn more about those here. However, the amplification chain comes with bothersome noise, against which the qubits must be isolated. In state-of-the-art systems, researchers solve these problems by incorporating additional microwave devices at the 15 mK stage, such as circulators and isolators, which route readout signals in a single direction, i.e., from input to output, and protect the qubits by blocking some of the noise coming from the readout chain, respectively. While this solution might enable high-fidelity qubit readout, the reliance on these large microwave components introduces a scalability limitation—which is an issue when the IBM Quantum team has their eyes set on a million-qubit quantum computer in the coming decades. In short, qubit readout chains urgently require innovative improvements. 

“The challenge with the current technology is that we use commercial magnetic isolators and circulators that do work well in cryogenic circumstances, but have the disadvantage of being big in size, expensive, and heavy. If you have a lot of them, because you would need to read out a lot of qubits, they will occupy a large volume in our dilution fridges,” said Baleegh Abdo, Master Inventor and Research Staff Member on the IBM Quantum team. Moreover, these isolators and circulators are magnetic, which can have adverse effects on superconducting qubits, further requiring that the quantum chips be placed inside magnetic shield cans, which inevitably degrade the readout fidelity due to the separation they impose between the various components in the readout chain. 

Fortunately, a team of IBM Yorktown and Almaden, led by Abdo, are inventing new on-chip devices to eliminate these big magnetic components. Josephson Directional Amplifiers (JDA) amplify signals in only one direction, replacing the need for combining circulators with JPAs,  and Josephson Isolator (JIS) devices work to block output noise in the microwave frequency band used for readout. Both of these Josephson junction-based directional devices are compatible with superconducting circuits, devoid of magnetic materials, and can be operated using a single microwave drive or pump.  They are constructed by using two Josephson Parametric Converters (JPCs), a Josephson Junction-based amplifier. Josephson junctions are lossless, nonlinear inductors named after Brian David Josephson, a Welsh theoretical physicist; If you want to learn more, check out this post.

“Our new solution uses technologies that allow for miniaturization while being compatible with the superconducting qubit technology,” said Abdo.

The team envisioned creating a readout motherboard that combines these Josephson Junction-based devices with other on-chip microwave components and wire bonding them to the printed circuit board. And, indeed, at the 2020 American Physical Society March Meeting, they successfully demonstrated the construction and testing of such a motherboard. 

However, since this motherboard was mainly a proof-of-concept, there’s still a considerable amount of work left before it’s ready for incorporation into an IBM Quantum device. Some of the questions which the team had to tackle are: How do we improve its performance? How do we integrate it, and how do we facilitate its packaging? 

The team provided some possible answers to these important questions at this year's APS March meeting (2021). 

A first improvement that the team made to both the JIS and JDA devices was getting rid of wire bonds. Leveraging IBM’s historical experience with bump bonds, the researchers produced JIS and JDA devices that they bump bonded directly to the PCBs. The bump bonds not only deliver incoming and outgoing signals, they also ground chip ports, carry the chip mechanically (you can think of these bonds as “glue”) and provide good impedance matching for broadband devices. The bump bond strategy is not brand new—they found use in the Quantum Hummingbird Processor, although in that case the quantum chip was not directly bump bonded to a PCB but to a silicon interposer, which, in turn, was bump bonded to a PCB.  A second improvement was increasing the JPC frequency (that is used to build the JIS and JDA devices) to better match the qubit readout frequency, requiring an update of the JPC design. Third, the team increased the strength of the couplings of the JPCs to the intermediate transmission line that is part of the JIS and JDA devices.

The tests showed great results: the JISs transmit signals in one direction with little added loss of about 0.5 dB when they are on versus off, whereas they attenuate signals in the opposite direction by more than 20 dB within a 10 MHz bandwidth. 

“We get a Josephson Isolator that allows the signal to propagate in one direction… and it blocks the signals that are propagating in the opposite direction. So, this is an important component that protects the qubit against noise coming from the output chain," said Abdo. 

Schematic of a JDA

 The JDA was also subject to improvements. In addition to being bump-bonded to the PCB in a similar manner as the JIS, the research team developed another breakthrough innovation: for the first time, they realized an on-chip 90-degree hybrid for the high-frequency pump needed for amplification. This achievement was made possible due to the use of the bump bonding providing good impedance matching for high frequency components. The new JDA device is capable of amplifying signals by 18 dB in one direction in a bandwidth of 3 MHz, whereas signals in the opposite direction are slightly attenuated. Here also, the insertion loss is limited to approximately 1 dB.

Core to these two directional devices is the JPC, a three-wave lossless mixing device operating at the quantum noise limit with two differential modes (often called “a” and “b”) that couple to a Josephson Ring Modulator or “JRM”. A JRM consists of four relatively large Josephson Junctions assembled in a loop and shunted by linear inductance. The differential modes are enabled by coupling two orthogonal microwave resonators having different resonance modes “a” and “b” to the JRM.

Apart from the differential modes, the JRM has a common mode that is driven off-resonance by the pump, which is a strong coherent microwave tone. The a-modes of the two JPCs in the JIS and JDA devices are coupled via a 90-dgree hybrid and the b-modes are coupled to an intermediate transmission line and 50 Ohm terminations. In the case of the JIS, the JPCs are operated in mixing or conversion mode, while in the case of the JDA, they are operated in amplification mode. Please look at the illustration figures in the sidebar. In short, JPCs which include the JRMs are used for building both the JIS and JDA devices, but the assembled devices operate in different ways. 

Schematic of a JIS

While all of these deliverables are a huge step forward in the journey towards building a universal fault tolerant quantum computer, further challenges are on the horizon, such as reducing the size of the motherboard and increasing the bandwidth of the JPCs so the JDA and JIS can support frequency multiplexed qubit readout. Multiplexing allows a single set of readout circuitry to measure a large number of qubits, which is key because the number of high-coherent qubits in quantum processors is expected to keep growing over the next years.

Abdo said: “In the long run, after we overcome both of these technological and research challenges, we believe we can use the motherboard device in much larger systems than 1000 qubits. And with our team we are currently designing JPCs that are more compact and have a larger bandwidth. We hope to get some preliminary results in the coming months."

Solving these issues successfully is of critical importance to scaling superconducting quantum hardware. Another interesting evolution is the possibility of moving from aluminum to niobium junctions, when building these directional devices.  

“Since our current JDA and JIS devices are based on aluminum junctions, they must be tested and operated at temperatures much lower than 1 K. At some point, it would be nice to build them using niobium junctions, which will make them more resilient, have higher saturation power, and easier to test and characterize,” said Abdo.  

This is just one of a few solutions to tackle the challenge of scaling up and miniaturizing superconducting quantum systems. IBM's team might converge on a solution like the one that Abdo’s team is working on, i.e., a motherboard which supports multiplexed qubit readout, which integrates multiple devices, or rely on state-of-the-art research into other components as a potential alternative.

Ultimately, the IBM team will draw on the best available research in the field when it comes to finding a solution to the problem of scaling up. As for the motherboard, down the road, it is likely that it will also host extra functional components, just like the classical version of the motherboard in our general-purpose computers of today.