The first time in the word history when a CEO invite a lot of other CEO to explain why I was wrong
NVIDIA helps advance autonomous cars but do not build cars
NVIDIA don't build robots but helps simulating them
NVIDIA don't build quantum computer but help simulating then also
NVIDIA will open a quantum research lab in Boston
Quantum computer from single atom and lazer beams. Atom can preserve quantum state for a very long time -> 1000s of qubits
Superconducting gate based quantum cmoputing : as Pascal Febvre. Gate speed at 10s of nano second. They get noise as CMOS. Now from 99 to 99.5 percent of accuracy fidelity. Open Modular approach. But not gooenough for production yet
Trap ion. QCCD : Quantum Charge Coupled Device. Extend QCCD at bigger scale. Hundreds on QBits
Neutral Atom technology. Scalability, thouthands of Qbits can be traped with Lazer Beams. A lot of progress in gate fidelity. Turn them from lab experiment to deliverable project. They delivered 4 machines. FOcusing on engineering
Traped Ions. They are used since 1995 because of the Atomic Clock. 30 years in this inverstment. Individual atom. Down at 0.2 nanometer. Tacked based room temperature. Distributed quantum computing. with laser beams. They have the best stability.
Superconducting. Use neealing technology not Gates. Easy to use, mush less sensitive for noise. Property to magnetic material => how to use that coputation to compute the hashing function of the blockchain. And they also can check the hash. Could be use to lower the power consumption needed to make the block chain work. In the public domain for over a year now.
What is the definition of usefullness in quantum computing ?
Quantum computers are a new scienific tool => huge potential to really advance the scienific frontier and make new discovery. This can start a new industry that noone could predict.
Application already in the aera in chemisty. How to generate Hydrogene without ? How to cool more efficiently ?
Quantum computer has a problem because everybody knows what a computer should be. Why not call them Quantum Engine ? a scientist instrument.
They are very complementary with CPU and GPU for specialise tasks
Of course, discovering new material implies a lot of computations
Improvement for medecine even with 16 Qbits. They will try to have 64 Qbit by the end of the year to solve chemisty problems.
Why nobody saw a more efficient approach and all other use the same ?
We have more commonlality than people can expect between all quantum technologies
What is the point to solve quantum problem with AI ?
These problems are impossible to solve today. Not for all problems, if the computation cost is too high.
To exdand the chemisty problems, quantum computing can help AI to train => GENQ-AI
Now it is about identifing to leverage the best of both world. To solve problems that are themself quantum, when classic methods do no work.
There will be no replacement but both quantum and classical computers will work together.
And the most part of a quantum computer is classic, except a small ship.
Quantum processing or quantum processor rather than quantum computer ?
Analog input and output. You want to solve as much as possible with the classic approach and only at the end to quantum computing
Agent AI quantum computing for next year ?!
Second panel :
Ben Bloom : Founder and CEO, Atom Computing
Matthew Kinsella : CEO, Infleqtion
John Levy : CEO and Co-Founder, SEEQC
Théau Peronnin : CEO and Co-Founder, Alice & Bob
Rob Schoelkopf : Chief Scientist and Co-Founder, Quantum Circuits
Pete Shadbolt : Co-Founder and Chief Scientific Officer, PsiQuantum
Quantum computer with neutral atoms. Very very lot a qubit. 1000s qubit. A lot of connectivity
Neutral atoms. Highy flexible technology. No need of a freezer but lazer. Atom => perfect qbits, perfect sensors and perfect clocks
Need to readout, integration => build digitaly control, multiplex quantum computer, on a ship. To be energy efficient. 3 nanoWatt to control one qbit. CPU, GPU, QPU
Cat Qbit => first correction inside the qbit. Shorten the timeline of development.
Superconducting. Error correction is the key to optain usefull quantum computation. Correct first then scale. New paradigm in Superconducting with error correction in the ship itself. Enhance fidelity we can get. A way to scale more efficently. Main challenge is to show error correction really works
Very large scale fault tolerant million qbits machine. Single photons. On a ship repurposing photon techology. Colling power. Breaking ground in the next few mounths at Chicago and Autralia.
Utility ?
Keep scaling. For sure we will need millions of qbit to change the world.
There are physical qbit and error corrected logical qbit. 10000 to 1, now more 100 to 1.
In last google paper each qbits needs 5 individual cables. So this will be a problem to scale to 1 million qbit
By the end of this decade.
We are learning how to build these machines.
The idea to create a space where we can explore is exactly the best we could expect.
Your are not limited by Moore's ? even if the Moore's law is not a law.
Scaling, what's enables you to scale at this speed ?
We can use classic computers to simulate.
The interface between CPU, GPU and actuators is not as good as it could be yet.
Scaling is not about puting more qbit but less hardware to check error and correction
A long coherence time is still micro second. Less than 1us latency to be able to correct before the next error shows up.
Millisecond is easy
Mircosecond is challenging
Nanosecond is hard
What qbit went wrong, how to adapt configuration to solve the problem
Latency does not scare me but latency and throughput do scare me.
NVIDIA provide large parallel computer to the world but started with games because it was easier to scale with it. Even if they wanted to do large scale computation.
NVIDIA didn't replace the computer, they added to it. So we did no arm because we added instead of replacing scalar computing with parallel computing
The whole will be better than the sum of the parts (CPU, GPU, QPU). But we need rational for that.
New algorithms will come.
Better error correction.
Thrid panel :
Krysta Svore : Technical Fellow, Microsoft
Simone Severini : General Manager, Quantum Technologies, AWS
Partner with quantum society. Topological qbit => encode qbit information non locally. Can simplify the amount of control needed.
Superconducting technologies with focus on error. Knowlege, speed and experience. They use microwaves. Some semiconductor experience can be translacted to superconducters.
50 qbit for chemisty and material science. Use the quantum computer to produce highly quality data.
How many logical qbit ?
April last year 4, then 12, then 28 with atom computing now. They work on 50 physical qbit.
The quality of error correction will tell which technology is best.
1 error to every 100 operation is not enough. 1 per quadrillion would be better.
