Sign in

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Matt McEwen

@matt.mcewen.xyz
308 followers 106 following 19 posts

Research Scientist with Google Quantum AI

PostsRepliesMedia
Matt McEwen @matt.mcewen.xyz · 20/12/2024
Each of these circuits present a new possibility to design an architecture around. That all three work well on a device not optimized for them proves we have what @craiggidney.bsky.social calls “circuit agility”, like cryptographic agility but for QEC implementation. Same QEC, different circuit!
180 8
Matt McEwen @matt.mcewen.xyz · 20/12/2024
It was also long assumed that you need to use CNOTs or CZ gates to do QEC. We demonstrate the surface code compiled to iSWAP gates, which is more natural for transmons and doesn’t couple to leakage states! Our device was optimized for CZs, the iSWAPS do suffer from a lot of stray CPHASE error…
150 5
Matt McEwen @matt.mcewen.xyz · 20/12/2024
We also demonstrate a walking circuit, switching which physical qubits we use to measure the syndromes. Physically, this lets us remove leakage on all qubits every 2 rounds. Logically, it moved the whole patch diagonally through space.
130 3
Matt McEwen @matt.mcewen.xyz · 20/12/2024
It’s long been assumed you need a square grid of nearest neighbour connections to do the surface code. Turns out three couplings is plenty! As in our recent work arxiv.org/abs/2408.13687, we show that scaling the distance for the hexagonal circuit from distance 3 to 5 halves the logical error.
150 5
Matt McEwen @matt.mcewen.xyz · 20/12/2024
These circuits are all based on the detecting region approach to QEC arxiv.org/abs/2302.02192 As long as these regions tile spacetime, the QEC works, which gives you huge freedom to modify the circuit. It's a treat to work on theory for something and then get to work on the experiment.
160 6
Matt McEwen @matt.mcewen.xyz · 20/12/2024
Circuits Galore! Check out our latest work demonstrating three new circuits for implementing the surface code: arxiv.org/abs/2412.14360 They all perform well below threshold on a device optimised for the standard surface code, but they each open up wild new possibilities for devices.
1286 28