Sign in

Zlatko Minev

@zlatko-minev.bsky.social
1.2K followers 151 following 727 posts

Google Quantum AI | Ex-Team Lead, IBM Quantum | MIT TR35 | Founder, Open Labs | Board, Yale Alumni Assoc | Yale PhD

PostsRepliesMedia
Zlatko Minev @zlatko-minev.bsky.social · 54m
A package mode keeps a very small fraction of its energy in the qubits' junctions below it. That fraction, the junction's energy participation, determines the coupling. Keep the qubit detuned so the eigenmode stays package-like; participation plus detuning give g, sign included.
Simulated vertical electric field magnitude of the second package mode over a 10x10 transmon array, with two lobes separated by a null line.
100
Zlatko Minev @zlatko-minev.bsky.social · 07/10/2026
Quantum error mitigation can be viewed as filter design. The filter controls how information paths contribute to the estimate. Its coefficients set the sampling overhead. Together with the reactivity, this connects information spreading, bias, and cost.
Tunable error cancellation: relative damping of Pauli paths versus weight for several tuning parameters, and the resulting sampling overhead.
110
Zlatko Minev @zlatko-minev.bsky.social · 06/10/2026
The most direct way to read a coupling: push two modes through each other and measure the gap. Replace the junction by a variable inductor and sweep it; at closest approach the branches are 2|g| apart. The sweep runs in a circuit tool on exported S-parameters.
Avoided crossing: two eigenfrequency branches versus qubit inductance approach each other and repel; the minimum gap equals 2|g|/2pi.
100
Zlatko Minev @zlatko-minev.bsky.social · 06/10/2026
Tomorrow the 2026 Physics Nobel is announced. A year ago: Clarke, Devoret, and Martinis, for macroscopic quantum tunneling and energy quantization in an electric circuit, the root of today's superconducting qubits. Michel was my PhD advisor. Any predictions this year?
260
Zlatko Minev @zlatko-minev.bsky.social · 05/10/2026
An observable can receive contributions from exponentially many quantum paths, yet have a much simpler response to noise. At finite precision, a compact effective reactivity can capture the structure needed for error mitigation.
Two example reactivity functions R(w) with their effective cumulant and multi-exponential descriptions, and the resulting noisy signal decay curves C(gamma).
110
Zlatko Minev @zlatko-minev.bsky.social · 05/10/2026
Different paths through a quantum circuit can share the same accumulated noise weight and respond identically to noise. Adding their signed contributions gives the reactivity R(w). Its Laplace transform gives the noisy signal in our noise-scaling model.
Paths through a layered noise lattice from initial state to observable, three highlighted with accumulated weights 2, 2 and 5; right, a signed histogram R(w) binning path contributions by accumulated weight.
130
Zlatko Minev @zlatko-minev.bsky.social · 04/10/2026
One full-wave eigenmode solve, one hundred couplings. Open every junction: the qubit resonances leave the package-mode band and the eigensolver sees only the package mode. The induced voltage on each port, with the port capacitance from electrostatics, gives g, sign included.
Heatmap of the signed coupling g/2pi in MHz for each qubit of a 10x10 array: positive on the left half, negative on the right half, near zero along the middle columns.
111
Zlatko Minev @zlatko-minev.bsky.social · 04/10/2026
A new postdoc fellowship worth passing along: the MIT Quantum Initiative (QMIT) fellowship, supported by the Moore Foundation. Up to 17 fellows, across physics, chemistry, materials, and the bio and Earth sciences. For PhDs from Jan 2025 on, starting in 2027. Due October 23.
140
Zlatko Minev @zlatko-minev.bsky.social · 03/10/2026
A superconducting processor sits in a metal box, and the box is a resonator. Load a 30 mm box with silicon and its fundamental lands near 6.5 GHz, inside the transmon band. One mode, shared by every qubit: a global decay channel and a long-range coupling nobody designed.
Sketch of a transmon qubit on a dielectric slab inside a package, showing the qubit's field circulating between its paddles and the package mode's mostly vertical field.
110
Zlatko Minev @zlatko-minev.bsky.social · 03/10/2026
Nice to see a theory prediction tested directly on hardware: chaotic quantum circuits show universal output statistics already at shallow depth, on a Google Quantum AI processor. Congrats to Arman Sauliere, Jacopo De Nardis, Andrea De Luca, Pedram Roushan and team.
141
Zlatko Minev @zlatko-minev.bsky.social · 02/10/2026
Four ways to compute a transmon's coupling to its package mode: different junction boundary conditions, different observables, two eigenmode solvers and two driven ones. On a 10x10 array they agree to 4.8% at worst, 2.9% on average. Pick the one that suits your workflow.
Bar chart comparing qubit-to-package-mode coupling for 25 qubits computed with four methods: EPR, avoided crossing, induced EMF, and impedance-matrix analysis. The four bars agree closely for every qubit.
100
Zlatko Minev @zlatko-minev.bsky.social · 02/10/2026
Not how many qubits, but what did we learn? That's the shift in DOE's new national quantum computing roadmap: useful science by 2028, then an open user facility.
110
Zlatko Minev @zlatko-minev.bsky.social · 01/10/2026
Same package mode, two components. Left: |Ez|, a smooth hump, textbook TM. Right: |Ex|, two lobes and a null line down the middle. An empty box has no right-hand picture. The silicon substrate gives the mode an in-plane component, and that is what the transmons couple to.
Simulated field magnitudes of the fundamental package mode over a 10x10 transmon array: vertical field Ez peaked at the center (left) and in-plane field Ex with two lobes and a null line down the middle (right).
100
Zlatko Minev @zlatko-minev.bsky.social · 30/09/2026
Very happy to share our new preprint, “Quantum error mitigation from information dynamics.” A circuit's noisy signal is the Laplace transform of one object, its reactivity. Error mitigation becomes filter design on it, with ZNE and PEC as limits of one tunable method.
Title and author block of the preprint 'Quantum error mitigation from information dynamics' with its Figure 1: (a) noisy expectation value vs noise rate, (b) the reactivity function, (c) cumulant, multi-exponential and tunable-error-cancellation mitigation, (d) accuracy vs cost for TEC, ZNE and PEC.
161
Zlatko Minev @zlatko-minev.bsky.social · 30/09/2026
Quantum jumps hold a special place for me. My PhD at Yale was about catching them mid-flight in a superconducting atom. Wonderful to see them now in sound: Stanford watched a single phonon leave a mechanical resonator in real time, read out by a superconducting qubit.
151
Zlatko Minev @zlatko-minev.bsky.social · 29/09/2026
A hundred transmons in a metal box, each colored by how strongly it couples to the box's lowest mode. New in Quantum Metal v0.9: our recent preprint as open tutorials you can run. Not a sign-off tool for a real device, but a surprisingly good way to learn.
3D plot of a 30 mm by 30 mm package with a 10 by 10 array of transmon paddle pairs, each colored by its coupling g/2π to the package's lowest mode (LSM110): blue (about -12 MHz) on one side, red (about +12 MHz) on the other, fading to white along the center line.
100
Zlatko Minev @zlatko-minev.bsky.social · 27/09/2026
Why does a horizontal transmon dipole couple to a vertical package mode? The silicon substrate modifies the mode: it develops an in-plane E-field at the qubit plane. Across a 10x10 array the coupling follows that field's cos-sin profile. Dipole estimate 14.2 MHz, simulated 15.6.
Left: sketch of a transmon qubit's in-plane electric field and the package mode's mostly vertical field. Right: extracted qubit-to-package-mode coupling versus position across five rows of a 10x10 array, all following one cosine-sine profile.
110
Zlatko Minev @zlatko-minev.bsky.social · 26/09/2026
Our new preprint is out: four ways to extract how strongly each qubit in a 100-qubit processor couples to the modes of its own package, benchmarked on a 10x10 transmon array. They agree to within 4.8%, and the spatial pattern has a clean dipole explanation. Fun one to work on.
Title and author list of the preprint 'Extracting Electromagnetic Bare Mode Couplings in Large Superconducting Quantum Processors' above an overview figure comparing four coupling-extraction methods: avoided crossing, energy participation ratio, induced EMF, and impedance-matrix analysis.
141
Zlatko Minev @zlatko-minev.bsky.social · 17/09/2026
Hiring alert: Our team at Google Quantum AI is hiring, see the post link below. It is a senior research scientist role, with ownership and leadership from concept to validated experiment. If you are finishing a PhD, please do not rule yourself out. www.google.com/about/caree...
careers.google.com
Senior Research Scientist, Workflow Tools Pathfinding, Quantum AI
As an organization, Google maintains a portfolio of research projects driven by fundamental research, new product innovation, product contribution and infrastructure goals, while providing individuals and teams the freedom to emphasize specific types of work. As a Research Scientist, you'll setup large-scale tests and deploy promising ideas quickly and broadly, managing deadlines and deliverables while applying the latest theories to develop new and improved products, processes, or technologies. From creating experiments and prototyping implementations to designing new architectures, our research scientists work on real-world problems that span the breadth of computer science, such as machine (and deep) learning, data mining, natural language processing, hardware and software performance analysis, improving compilers for mobile platforms, as well as core search and much more. As a Research Scientist, you'll also actively contribute to the wider research community by sharing and publish
173
Zlatko Minev @zlatko-minev.bsky.social · 15/09/2026
A delight to speak this morning at the ASQC workshop at IEEE Quantum Week in Toronto — an overview of quantum hardware design: energy, circuits, and Metal. Grateful to all who built the open tools that take a layout to a Hamiltonian. A real gap remains; we close it together.
Left: banner of the Advanced Simulations of Quantum Computations (ASQC) workshop, IEEE Quantum Week 2026, Toronto, September 14-15. Right: title slide of Zlatko Minev's talk, Overview of Quantum Hardware Design: Energy, Circuits, and Metal, showing a Bloch sphere above a chip schematic and a Quantum Metal layout.
120
Zlatko Minev @zlatko-minev.bsky.social · 15/09/2026
If you teach quantum computing: which idea do your students get stuck on hardest?
000
Zlatko Minev @zlatko-minev.bsky.social · 08/09/2026
You can't switch a device's noise off. But you can turn it up. Measure at several noise levels, then extrapolate back to the zero you can never reach. A straight line lands at 0.936. The exponential lands exactly on 1.000 — because this decay happens to be exponential.
100
Zlatko Minev @zlatko-minev.bsky.social · 03/09/2026
A qubit idling in a slowly drifting field loses its phase. Flip it halfway through and the drift undoes itself. Here that's 34.5% fidelity becoming 100%, at any idle time. The catch is hiding in one word: the drift has to be the same in both halves.
220
Zlatko Minev @zlatko-minev.bsky.social · 01/09/2026
What's the one explainer about quantum noise or error mitigation you wish existed?
000
Zlatko Minev @zlatko-minev.bsky.social · 28/08/2026
A T1 on a calibration report isn't a deadline. It's the 1/e point. Wait one T1 and 36.8% of the excited population is still there. Wait three and you're near 5%. T2 is the second clock: how long the phase lasts.
220
Zlatko Minev @zlatko-minev.bsky.social · 26/08/2026
Readout error across many qubits is correlated, and the matrix you'd need to invert grows exponentially. Randomize the bits before measuring, undo the flip in software, and the mess collapses into a single scale factor you can divide out. That's TREX.
100
Zlatko Minev @zlatko-minev.bsky.social · 24/08/2026
This is what noise looks like in raw data. A 3-qubit GHZ state should give only 000 and 111. On hardware you also get six bitstrings that shouldn't exist — and 111 has decayed further than 000. Noise doesn't throw an error. It just moves counts.
100
Zlatko Minev @zlatko-minev.bsky.social · 21/08/2026
A qubit doesn't just get pushed off course. It shrinks. Coherent errors rotate the arrow. Incoherent noise pulls it inward, toward the middle of the sphere — energy relaxation and dephasing. Rotations you can undo. Lost information you can't.
111
Zlatko Minev @zlatko-minev.bsky.social · 20/08/2026
Left: six fluxonium qubits laid out in Quantum Metal. Right: the same chip, fabricated and wire-bonded. Nice work from Figen Yilmaz, Christian Kraglund Andersen and colleagues at QuTech, who extended EPR analysis to very anharmonic circuits and checked it against measurement.
100
Zlatko Minev @zlatko-minev.bsky.social · 19/08/2026
You can't start in a perfect |0> - and it compounds. State purity Tr[rho^2] falls exponentially with qubit count. At 1% error per qubit, about a third survives at 50 qubits. At 5%, essentially none.
110
Zlatko Minev @zlatko-minev.bsky.social · 17/08/2026
A qubit holds a continuum. You get one bit out. Projection noise — and a perfect device does it too. The spread is largest at p = 0.5: the least committed state is the noisiest to read.
110
Zlatko Minev @zlatko-minev.bsky.social · 13/08/2026
What is a qubit, really? Not a tiny ball. Not a 0 or 1 waiting to be read. It's an arrow on a sphere — two angles fix it completely. Every noise process you meet later is a story about that arrow.
232
Zlatko Minev @zlatko-minev.bsky.social · 11/08/2026
If you want to try Quantum Metal without installing anything, you can now — 90+ tutorials have a Colab and Binder button, so you click one and you're designing a chip in a live notebook. Mostly there to help students and newcomers get started.
131
Zlatko Minev @zlatko-minev.bsky.social · 10/08/2026
Sometimes the qubit is right and the measurement lies. A |1⟩ gets read as 0, a |0⟩ as 1 — so the probability you measure is biased toward the middle. The good news: readout error is measurable, and you can correct much of it back out.
110
Zlatko Minev @zlatko-minev.bsky.social · 06/08/2026
When you first ran a circuit on real quantum hardware, what surprised you? Which noise source hit you first — and which took longest to actually understand? I'll compile the replies and share them back.
310
Zlatko Minev @zlatko-minev.bsky.social · 04/08/2026
A tiny gate error is a big deal because it's coherent — it adds up. Rotate a qubit 3° too far on every X gate. Invisible once. But after ~30 gates it's a 90° error, and by ~60 the answer has fully flipped. The circuit is now confidently wrong.
111
Zlatko Minev @zlatko-minev.bsky.social · 31/07/2026
A quantum computer doesn't give you a probability — it gives you a coin. One run, one bit. To learn the probability you average many shots, and that estimate has spread: sampling noise. Want half the spread? Run 4× as many (it shrinks like 1/√N).
240
Zlatko Minev @zlatko-minev.bsky.social · 30/07/2026
Before you can simulate a quantum chip, you first turn it into a mesh — a scaffolding of little tetrahedra the fields get solved on. Nice thing about the open tooling: it renders straight to Gmsh, so an open-source design can go the whole way to an open-source solve.
131
Zlatko Minev @zlatko-minev.bsky.social · 29/07/2026
Teaching an intro to circuit QED at the Quantum Device Workshop reminded me: explaining the fundamentals forces you to actually understand them again. You can't hand-wave in front of people seeing it for the first time.
130
Zlatko Minev @zlatko-minev.bsky.social · 28/07/2026
Someone in the community asked if Quantum Metal could add a SNAIL — three Josephson junctions on one arm, one on the other, the element behind Kerr-free parametric amplifiers and three-wave mixers. The maintainers built it. Physics from Frattini, Sivak, Devoret.
500
Zlatko Minev @zlatko-minev.bsky.social · 27/07/2026
Correction and mitigation aren't rivals. Correct errors as you go with redundancy; or let them happen and subtract their effect afterward. And mitigation isn't just a NISQ stopgap — we'll always run the biggest circuits the hardware allows, so there's always error to clean up.
110
Zlatko Minev @zlatko-minev.bsky.social · 24/07/2026
Half of laying out a superconducting chip is really just routing — every line reaching its target at the right length without crossing another. Tedious by hand. Nice that the open tooling can take some of it off your plate: meanders for length, mixed routes to thread through.
100
Zlatko Minev @zlatko-minev.bsky.social · 23/07/2026
One reason a quantum computer doesn't do what the math says: ask it to flip a qubit and the real pulse rotates a hair over or under 180°. Barely visible once — but across a long circuit those consistent misses pile into a wrong answer. That's coherent noise.
130
Zlatko Minev @zlatko-minev.bsky.social · 23/07/2026
Good to see the DOE's Quantum Genesis awards land across so much of the quantum community — congratulations to everyone selected.
110
Zlatko Minev @zlatko-minev.bsky.social · 23/07/2026
A quantum computer that learns from its own errors — and never stops to recalibrate. Our team at Google Quantum AI, in Nature: an RL agent steers thousands of controls from the live error-correction signal, cutting logical error ~20%. Vlad Sivak, Paul Klimov + team.
110
Zlatko Minev @zlatko-minev.bsky.social · 22/07/2026
For a while the community had been asking Quantum Metal for airbridges — small crossovers that hop over a CPW to tie the ground plane together and quiet the slotline modes that spoil a qubit. They're in there now, thanks to Clark Miyamoto and the QDC maintainers who built it.
100
Zlatko Minev @zlatko-minev.bsky.social · 21/07/2026
It was a pleasure to meet Stephen M. R. Covey, who visited Google to speak about trust. His point stuck: trust isn't a soft virtue, it's economic. Low trust slows everything down. Almost no research runs on enforcement — it runs on belief. Where has trust changed your speed?
000
Zlatko Minev @zlatko-minev.bsky.social · 20/07/2026
Google has two open research calls: early fault-tolerant algorithms, and securing the quantum computer itself — side channels, control pulses, user isolation. Not post-quantum crypto. Up to $100k, closes Aug 7. research.google/programs-an...
research.google
Google Academic Research Awards
062
Zlatko Minev @zlatko-minev.bsky.social · 21/06/2026
Quantum Metal needed a desktop GUI in 2020. I couldn't see chip design happening without one. In 2026 the GUI is optional — Colab, CI pipelines, AI agents all want headless. Same project, different field.
110
Zlatko Minev @zlatko-minev.bsky.social · 19/06/2026
Back from QDW 2026. What I'll remember: how many groups had independent Quantum Metal workflows I'd never seen. Noted for the project: better open-FEM docs (gmsh + Elmer + Palace), more beginner tutorials, a clearer contribute path. What would you add?
000