Hacker News Reader

Comment by IIAOPSW | Hacker News Reader

A bit misleading. Waveguides in Silicon have been used to make photon qubit gates for a while now. It isn't the first time Silicon has been used.

The real takeaway is "Here we present a two-qubit logic gate, which uses single spins in isotopically enriched silicon".

Paper found here: http://www.nature.com/nature/journal/vaop/ncurrent/pdf/natur...

Using spin qubits is nothing new, but normally its done in an ion trap. Doing it in Silicon is novel. I've barely skimmed the paper so I don't really know what their decoherence properties are.

View on HN

You seem to know a lot more about this than I do: in your opinion how much of a leap is this? Could this be the Chicago Pile of quantum computing or just another D-Wave?

So I did ctrl+f for "coherence time" and found this:

A marked improvement in coherence times has been observed by defining the quantum dots in silicon, which can be isotopically purified, such that quantum dots with single-spin fidelities above the threshold of surface codes can be realized

Meaning they get a bit less than a 10% error per operation.

the thing that will make quantum computing practical is quantum error correction. I describe it as "the hospital problem". Suppose there's a car accident. An ambulance gets sent to pick up the survivors and fix it. But an ambulance is also a vehicle and so there is a chance it will also get in an accident. You can send more ambulances to fix the situation, but those ambulances are also vehicles and could also get into accidents. The layout of the roads, frequency of hospitals, and probability of accident per meter will determine on average how many problems get fixed per ambulance. If on average an ambulance gets into less than one accident then it is worth it. Otherwise ambulances cause more accidents than they fix. The point where ambulances fix exactly as many accidents as they create is called the threshold. Once you pass the threshold, with sufficient hospitals you can make the expected number of accidents arbitrarily low (aka the probability of successful computation arbitrarily high).

When people talk about surface codes, they mean the broad class of error correction schemes which look a bit like the toric code. The toric code has a threshold value around 10% which is great! However it is literally a lattice of atoms (or other spin systems) embedded on a torus. Construction is difficult. It also doesn't scale. A toric code with a few dozen qubits works as well as one with a few hundred. In fact several no-go theorems show that any transitionally invariant lattice in less than four dimensions won't scale. My research is actually in getting around these theorems by throwing out the lattice idea entirely.

This research is important, but the real Chicago Pile of QC will be passing the threshold of an easy to implement correction scheme.

It's information and analogies like this that people should be reading - I'm definitely far more informed about QC now. Thanks for taking the time.

"When people talk about surface codes, they mean the broad class of error correction schemes which look a bit like the toric code. The toric code has a threshold value around 10% which is great! However it is literally a lattice of atoms (or other spin systems) embedded on a torus. Construction is difficult. It also doesn't scale. A toric code with a few dozen qubits works as well as one with a few hundred. In fact several no-go theorems show that any transitionally invariant lattice in less than four dimensions won't scale. My research is actually in getting around these theorems by throwing out the lattice idea entirely."

I'm not sure what you are trying to say here, but it doesn't seem to be standard thought.

When you lookup "surface code" the first thing that comes up is "toric code". Off hand there's the toric code, the 3d toric code, welded toric code, color codes, and the surface code itself (which is a lattice of "measure qubits" next to a lattice of "data qubits"). The basic idea on all of them is the same as Kitaev's original insight of topologically protected data. IMO its completely fair to lump these together as "surface codes".

If you disagree with my choice of words then I'm happy to concede and rephrase. If something else I said doesn't sound kosher please by all means lmk.

No, he's correct. In addition I would add that passive error correction is taking leaps and bounds lately, making dephasing essentially a non issue in multi-qubit systems (scalability is yet to be fully manifested)

correct in what respect? I'm not trying to be rude here but it is unclear what exactly I'm being called out on. Do you disagree with my use of the word "surface codes" or do you disagree with the way I've lumped a bunch of QEC's together under one name? Or is the disagreement something else entirely.

And yeah I'm aware of the advances in passive QEC. Exciting time to be alive!

The he I meant was you, I'm not calling you out.

I assume you're talking about the brothers Horodecki's no go theorems?

Also Bravyi and Tehal. And a few others that show up in this review paper. http://arxiv.org/pdf/1411.6643v1.pdf

Does doing this in silicon mean we may start to see mass produced qubit ICs? You'd have to keep coherence between all of the quantum circuit elements to actually have a calculation carried out properly, wouldn't you?

Perhaps this could lead to some sort of quantum FPGA equivalent within the next few years?

The answer is a solid maybe.

Like I said upthread, it all depends on quantum error correction. Right now they are below the threshold for some surface codes (note: in this context below a threshold is a good thing). However surface codes require a torus topology, and existing chip technology generally flat. They are above the threshold for topologically simple codes like Stein and Shor thus ruling them out. While their research is a step forward, what they have in their lab right this instant cannot be made into a scaleable quantum computer.

Interestingly enough, most quantum computers (both in theory and practice) are more like fpga's than like a classical CPU architecture. The general purpose qubits just kind of sit there and a quantum circuit is implemented on them with laser pulses. The only exception I'm aware of is the silicon waveguides group I mentioned up top.

Here's how I have understood QC and it's current state of engineering: A 2-qbit gate can do 2-qbit problems. A pair of 2-qbit gates can do 2 2-qbit problems at a time, not a 4-qbit problem. For Shor's Algorithm, you need n-qbit gates, where n is the size of your problem. Even if the claims of the paper are true, making n-qbit gates out of silicone is just as hard as making n-qbit gates the way they are done in physics labs.

Is my understanding of the phenomena largely within reality?

What you are interested in knowing about is Universal Quantum Gates. These are the quantum analog of AND, OR, NOT from which all other computations can be built.

https://en.wikipedia.org/wiki/Quantum_gate#Universal_quantum...

In answer to your question, you are mistaken. The Cnot gate plus a few Single qubit gates {haddamard, pi/8, phase shift} are sufficent for any quantum algorithm including shors.

Also not to be pedantic but computers are made of silicon. Breasts are made of silicone.

The difference in e is then from electrified to electrifying?

Reply on news.ycombinator.com