11 October 2023
Controlling Qubits at Scale
In our newest paper we demonstrate a way of controlling qubits that can achieve low gate error rates, parallel qubit control & exclusive use of technologies that scale.
Designing architectures for quantum computing is hard because three things must be achieved:
Extremely low gate error rates that stay low as the number of qubits increases.
Parallel qubit control - the ability to perform different operations on all of the qubits at the same time*.
The exclusive use of technologies that can be built at scale and integrated into a single device.
There are many architectures which solve one or two of these challenges but none has yet achieved all three simultaneously. In our newest paper, published in Physical Review Letters, we demonstrate a way of controlling qubits that can.
We use a new technique, Forced-Motion Addressing, which makes it possible for large numbers of qubits to be controlled in parallel and with negligible error rates while using only one high-frequency qubit control line. It does this using simple control structures that can be integrated into chips built at scale on standard semiconductor production lines.
Here’s how it works.
The parallel control challenge
Qubits are controlled using high-frequency (either microwave or laser) signals. Most architectures approach the challenge of parallelising qubit control by supplying independent high-frequency signals to each qubit.
This approach works well for small processors where non-scalable techniques can be used, such as connecting one coaxial cable to the chip per qubit or shooting lasers over the surface of the chip. However, it becomes extremely challenging to make this approach work at scale and with low error rates for two reasons.
First: crosstalk. Interference between all the high-frequency signals makes it hard to keep errors low as the number of qubits increases.
Second, it is incredibly hard to build chips which implement these architectures at scale:
Just connecting thousands of microwave or optical signals to a chip is beyond the cutting edge of existing packaging technology - and that’s before one starts thinking about where those signals come from and how they get to the chip.
How does one fit all of the structures required to deliver these signals to the qubits into a crowded chip while managing cross-talk?
The power dissipated by all of these high-frequency signals needs to be managed and removed from the chips.
We need another way.
(a) A typical parallel qubit control architecture, with independent control structures and sources for each qubit. (b) The Forced-Motion Addressing architecture, where the qubits are controlled using a single high-frequency line combined with localised low-frequency electric fields.
Our approach introduces a new paradigm for qubit control, allowing a single high-frequency structure to control large numbers of qubits in parallel. It does this by leveraging another form of parallel control that’s already built into ion trap chips but not normally used for the qubit states: the electric fields used to trap ions and move them around the chip.
These electric fields are created by applying low-frequency voltages to trapping electrodes built into the ion-trap chip. These electrodes are engineered to allow individual ions to be moved around in parallel and with low cross-talk.
Crucially, the fact that these signals are low frequency and produced by small voltages makes them vastly easier to integrate at scale and with low cross-talk than the high-frequency qubit control signals. They dissipate almost no power and, thanks to our WISE architecture, we know how to build and wire them up at scale.
To date no one has used trapping electrodes for qubit control because ion qubits respond to magnetic not electric fields. This is where our new technique comes in: we use the trapping electrodes like switches to locally control the interaction between each qubit and a single, shared high-frequency line. As a result, even though all qubits see the same control field, their interaction with it is controlled by the electric fields.
The technique works by applying small (millivolt) oscillating voltages to the trapping electrodes to make the ions vibrate backwards and forwards inside the trap. These vibrations change how the ions respond to the high-frequency control field. As a result, the trapping electrodes can provide fully localised control of the qubits, despite the high-frequency qubit-control field being shared between all qubits.
Forced-Motion Addressing lets a single high-frequency line do all the heavy lifting for an entire chip, with small localised voltages telling each ion what gates to do in response. By reusing the transport electrodes for qubit control, it lets us achieve full parallel control while adding only minimal extra complexity to our chips.
Integrating large numbers of voltages into chips while managing interference is a well-solved problem in the microelectronics industry, and one that has already been solved to make the ion traps work.
If you’re interested to go deeper into how the technology works, read our paper, Coherent Control of Trapped-Ion Qubits with Localized Electric Fields
Built for scale
The real test of a quantum architecture is whether big computers are made from “more of the same” stuff that small computers are made of. Afterall, the purpose of building today’s small-scale quantum computers is to pave the way for tomorrow’s quantum supercomputers, but this only works if they’re built the same way and reach the same error rates.
Ion traps are the leading approach to quantum computing and have some of the most mature architectures around. Yet, large numbers of high-power, free-space laser beams and other non-scalable technologies still play critical roles in roadmaps. Not ours! A focus on building technology that scales is the heart of everything we do at Oxford Ionics.
*There is some nuance about exactly what one means by “at the same time”. Long story short, if the time it takes to perform a gate on each qubit increases linearly or worse with the number of qubits, it doesn’t count as parallel.