Qubits
Transmon and fluxonium.
Two designs for the same job. We characterize both: how long they hold their state, how cleanly they can be driven, and what limits them.
Qubits on the bench. AI on the desk.
Two technologies are being built at the same time, and we work on both with our own hands: superconducting quantum circuits, in experiment and in theory, and AI systems, built and used every day. This page is where we explore them, explain them and ask where they meet.
- StatusHands-on and ongoing
- HardwareTransmon and fluxonium qubits
- AIAgents, memory and local models
Play with it
Drive a qubit.
A qubit is not a switch that is on or off. Push it with a microwave tone and it swings smoothly between 0 and 1. Change how hard you push, and how far off its natural frequency you are.
This is the textbook two-level atom under a steady drive, drawn live. It is not a result of ours and not data from any device.
You got ? · 1 came up 0 of 0 times
- Highest chance of reaching 1
- 100 %
- Swing rate, in units of the drive at 1
- 1.00
Where a qubit lives
Colder than deep space.
Each plate of the refrigerator is colder than the one above it. The last sits about a hundredth of a degree above absolute zero, some 270 times colder than the space between the stars. Only there does the circuit stay quantum long enough to use.
On the bench
Circuits that behave like atoms.
A superconducting qubit is an electrical circuit, cooled until it stops behaving like one. We work on both sides of it. In experiment: making the circuits, cooling them close to absolute zero, and measuring what they do. In theory: modelling them, and working out what to build next.
Readout
A qubit meets a resonator.
A qubit is read through a microwave resonator coupled to it. Drive that pair hard and its spectrum splits and reshapes in ways that say a great deal about both. We study that regime in experiment and in simulation.
Algorithms
What is worth running.
Real qubits are noisy. We study which quantum algorithms still make sense on hardware as it actually is, rather than as it is imagined.
Method
Measure, simulate, compare.
Every measurement is checked against a simulation of the same circuit. When the two disagree, that is where the physics is.
Further out
Two directions we are keeping open. Neither is decided, and we will say more when there is more to say.
- Hybrid platforms. Joining superconducting circuits to other quantum systems, so each does what it is best at.
- Topological superconducting qubits. Qubits protected by the shape of their quantum state rather than by careful isolation. A long-term goal.
Under the coldest plate, where the samples are mounted. 
Wiring a sample holder by hand. 
The finished holder, before it goes into the cold.
The laboratory is in Taiwan and is not owned or operated by Vishwa Vigyan. Measurements made there are shared only after publication and the relevant approvals, so this page shows directions and equipment, not results.
Play with it
Two ways to build a qubit.
Thread a magnetic field through each circuit and watch its energy levels. The transmon ignores it. The fluxonium reshapes itself, and at the half-way point its two lowest levels come close together, far from all the others.
Both circuits are solved live from their standard equations, with illustrative values (transmon: Ej 15, Ec 0.25; fluxonium: Ej 4, Ec 1, El 1, all in GHz). These are not our devices and not a result of ours.
- Transmon qubit frequency
- 5.21 GHz
- Fluxonium qubit frequency
- 0.58 GHz
- Fluxonium: next gap is this many times larger
- 5.8
On the desk
Not one AI. A team of them.
One model answering one question is the old picture. The new one is many agents, each with a job, sharing one memory: one plans, others build, one checks, one ships. Give them a task and watch it move.
An illustration of how such a team is organised, not a recording of a real run.
Waiting for a task.
What we have, what we are after
AI that does the work.
We do not only read about AI. We build systems with it and run our own work on them, then sell what proves itself.
Running today
- A working team of AI agents with a shared memory, used every day for research notes, writing, code and these websites.
- QuoteDesk and the custom systems behind VV Automate: AI that reads a customer’s enquiry and returns a priced quotation.
- A working AI prototype for a wellness company, built in days.
- AI in the research loop: writing simulation code and analysing measurement data alongside the physicist.
Building toward
- A swarm: many specialised agents that plan, build and check each other’s work with little supervision.
- Local-first AI: models and memory that run on your own machines, so private work stays private.
- AI in the lab itself: reading each measurement as it arrives, proposing the next one, keeping the logbook.
- The far goal: an AI that learns by itself from its own work, a step on the road to general intelligence.
I can help you build your AI.
For your business
The same tools, working for you.
What we build for ourselves here, we build for businesses at VV Automate: AI agents, bots and custom systems that take the repeated work off your team.
Field notes · October 2026
Where the two fields are heading.
Short notes, kept current. What is established is stated plainly and linked to its source. What is opinion is ours.
- QuantumError correction has crossed its threshold.For years, adding qubits to an error-correcting code made things worse. In 2024 a superconducting processor showed the opposite: each step up in code size roughly halved the logical error rate. The road from there to a useful machine is long, but it now points downhill.Acharya et al., Nature 638 (2025)
- QuantumThe transmon is no longer the only serious qubit.The transmon is simple and forgiving, which is why most processors use it. The fluxonium is harder to build and control, but it keeps its quantum state for longer: coherence above a millisecond has been reported. Which trade wins at scale is not settled.Somoroff et al., 2021
- QuantumThe hard problems are now engineering problems.Wiring thousands of lines into a refrigerator, keeping every qubit calibrated, and removing the microscopic defects that drain energy from circuits. Progress here is slow, unglamorous and decisive.
- AIFrom answers to work.The shift of the last two years is from models that answer a question to systems that carry out a task: read files, run code, use tools, check the result and try again. What limits them now is less the model than what surrounds it: memory, context, and how they are supervised.
- AISmall and local is getting good enough.Models that run on a laptop cannot match the largest ones, but for many jobs they do not need to. Private data, low cost and no waiting are reasons to keep work on your own machine, and a strong reason to learn how these systems are built.
- Where they meetAI is already useful to quantum hardware.A neural network has decoded the errors of a real quantum processor more accurately than the best hand-written decoders. The same kind of learning is being pointed at calibration and control. This direction is paying off today.Bausch et al., Nature 635 (2024)
- Where they meetQuantum for AI is still a question.Nobody has yet shown a quantum computer doing a practical machine-learning job better than an ordinary one. Loading real data into a quantum state is expensive, and training often stalls. It may change. We treat it as an open question, not a product.
Coming: the playground
Learn it by playing with it.
The two pictures above are the first pieces of something larger: a place where anyone can learn quantum computing and AI from scratch, by trying things and seeing what happens.
Being built
Quantum playground.
From a single qubit to gates, circuits and noise, each step something you can move with your own hands. No equations needed to start.
Being built
AI playground.
What a model is, how it learns, how an agent uses tools and memory, built up from small working pieces you can change and break.
Under construction. Until it opens, students can start here:
- PlayUse the two pictures on this page until the behaviour feels obvious.
- Build smallSimulate one qubit in Python, or wire one AI agent to one tool.
- Read one sourcePick a field note above and read the paper behind it.
- AskWrite to us with what you built and where you are stuck. Guidance is free.
Projects