How Do You Build A Quantum Computer?

[VictorMedvil]

So I have heard that IBM and Google have constructed Quantum computers, how does one actually construct a Quantum computer? I would like as much detail as possible on the construction of one of these devices and how it works exactly, I wanted to buy a Spin Q Quantum computer and I understand how binary computers work however not so much for Quantum computers could you guys give a insight into the process of making one?

 

[anorlunda]

This article may help you understand.

https://www.technologyreview.com/2020/02/26/916744/quantum-computer-race-ibm-google/

I hope your budget is as big as IBM's. Just kidding. Seriously, it is still on the leading edge of technology that only a few researchers can achieve. It is far from the reach of DIY.

 

[VictorMedvil]

This article may help you understand.

https://www.technologyreview.com/2020/02/26/916744/quantum-computer-race-ibm-google/
View attachment 285100I hope your budget is as big as IBM's. Just kidding. Seriously, it is still on the leading edge of technology that only a few researchers can achieve. It is far from the reach of DIY.

Well no, my budget is not 1 trillion dollars like Google and IBM however I would still like to understand how to build a Quantum computer, I have built many binary computers and would like to add a quantum computer into the mix.

 

[f95toli]

Well no, my budget is not 1 trillion dollars like Google and IBM however I would still like to understand how to build a Quantum computer, I have built many binary computers and would like to add a quantum computer into the mix.

There are many ways to build a quantum computer. There are platforms built around superconducting qubits, spin qubits, ion traps and photonics; just to mentioned a few.
One thing they all have in common is that they require a LOT of expensive kit to operate. The actual "heart" of the computers is not actually that complicated for some of these implementation (say superconducting or spin based ones) and if they were mass produced (which they are not, you can't buy one) they would probably not be very expensive. However, the computers require a LOT of microwave equipment ( for superconducting and spin based platforms) or lasers (ion traps or photonics) since all signals in and out are in the microwave range (typically 4-8 GHz). Moreover, all platforms with the exception for ion traps also need to be cooled down to very low temperatures; the type used by IBM and Google are cooled below 20 mK.
Currently, the equipment you need for a "basic" quantum computer with a handful of qubits will cost you about $1-2M . You then also need a lot of custom HW and a LOT of software to get it to actually do anything.

It is quite unlikely that we will ever get to the point where you can buy a stand-alone quantum computer and put it on your desk; even if the technology somehow develops to the point where it would be possible there simply wouldn't be much point; cloud access to a HPC system which includes a quantum processors is likely to make much more sense.
It is the same reason why very few people currently have a TPU of a FPGA accelerator card on their desk; they make much more sense when integrated with a HPC system

(yes, I am aware that there are already companies making very simply processors that do (nearly) fit on a desk; but they are meant for demonstrations and teaching; they are not full-scale processors; and you still need to supporting electronics and SW

 

[pbuk]

It is quite unlikely that we will ever get to the point where you can buy a stand-alone quantum computer and put it on your desk

This.

However you can write software using QC algorithms and run it on a 'regular' computer using a QC simulator.

 

[Windadct]

one piece at a time...

 

[anuttarasammyak]

Hello.　The latest video　

Quantum Hardware Design: Energy, Circuits, and Metal | Qiskit Seminar Series with Zlatko Minev​

1,885 views
Streamed live on Jun 26, 2021

could be informative.

 

[pbuk]

Or you could https://quantumai.google/quantum-computing-service.

 

[.Scott]

could be informative.

I and others in these forums have made the important point that "quantum computers" are specialized processors that would not be used to drive the user interface. A user would use conventional computers (not necessarily high-speed) to access these quantum devices. Even in embedded operations, there would either be no user interface or one provided by a conventional computer or microprocessor.

But the video that @anuttarasammyak has linked to makes another important point - that these devices are more in the category of "quantum circuitry" than "quantum computers". They can be made programmable, but for efficiency and performance, most are likely to be dedicated specialized information processing units.
The "programming" described in the video is a lot more like what you do to get a custom MMIC (microwave circuit) working than what it takes to get either programmable logic (like a gate array) or an embedded processor working.

The biggest thing tying quantum computing devices to the lab (or keeping them cloud-based) is their operating temperatures which are currently in the 10-100mK range.

 

[f95toli]

I and others in these forums have made the important point that "quantum computers" are specialized processors that would not be used to drive the user interface. A user would use conventional computers (not necessarily high-speed) to access these quantum devices. Even in embedded operations, there would either be no user interface or one provided by a conventional computer or microprocessor.

But the video that @anuttarasammyak has linked to makes another important point - that these devices are more in the category of "quantum circuitry" than "quantum computers". They can be made programmable, but for efficiency and performance, most are likely to be dedicated specialized information processing units.
The "programming" described in the video is a lot more like what you do to get a custom MMIC (microwave circuit) working than what it takes to get either programmable logic (like a gate array) or an embedded processor working.

The biggest thing tying quantum computing devices to the lab (or keeping them cloud-based) is their operating temperatures which are currently in the 10-100mK range.

The biggest thing tying quantum computing devices to the lab (or keeping them cloud-based) is their operating temperatures which are currently in the 10-100mK range.

A couple of points.
Here, firstly a "circuit" in quantum computing is essentially just another name for "algorithm" and the words are used interchangeably. In the current (NISQ) era running a circuit essentially means initialising the circuit to the same state every time and then running exactly the same thing many, many times. Note that the "circuit" is not "hard coded" into the the actual processor in any way; although some designs are more suitable to certain classes of circuits/algorithms (because of e.g. topology/connectivity)
However, in the future when we fully error corrected machines this will change and we will have quantum processors that are programmable "on the fly" meaning you can implement feedback in a way that is more similar to what you do with conventional logic (with the usual constraints related to measurements enforced by the no-cloning theorem etc). That said, even then they will s be used as co-processor to conventional computer; using a quantum processors as a "general purpose" processors does not make sense.

Secondly. the reason for why quantum computers are still (mainly) used in the lab is NOT because you need to cool them down. Note that only some platforms (e.g. superconducting and spin) platforms require mK temperatures; e.g. ion traps and photonic platforms do not (they sometimes use 4K cryogenics but that is very easy, we can cool whole particle accelerators to 4K)

Moreover, even for the platforms that DO require mK temperatures the cryogenics is an engineering challenge but it is certainly now a showstopper even today. If someone had a viable design for a processor with say a million qubits we could most likely build that today; all you need is a bigger fridge (running multiple dilution units in parallel). Such designs have been around for several years, they are relatively expensive to build (a few million dollars) but compared to the cost of a full system that is insignificant.
In fact, IBM has already started building such a system (look up "GoldenEye") which is supposed to be ready in a couple of years.

 

[.Scott]

... in the future when we fully error corrected machines this will change and we will have quantum processors that are programmable "on the fly" meaning you can implement feedback in a way that is more similar to what you do with conventional logic (with the usual constraints related to measurements enforced by the no-cloning theorem etc). That said, even then they will s be used as co-processor to conventional computer; using a quantum processors as a "general purpose" processors does not make sense.

Secondly. the reason for why quantum computers are still (mainly) used in the lab is NOT because you need to cool them down. Note that only some platforms (e.g. superconducting and spin) platforms require mK temperatures; e.g. ion traps and photonic platforms do not (they sometimes use 4K cryogenics but that is very easy, we can cool whole particle accelerators to 4K)

So for an application that has an inherent portability requirement - for example, if I needed quantum digital processing to help identify or filter targets for a missile - would you say that we are already pretty close to a solution? Say within a decade?

Also, you can see that in an application of this sort, you would still need to front end the quantum device with conventional electronics - but not necessarily a full computer processor?

Also:
When I say "lab", I am referring to the alternative to my office. My computer, logic scope, small power supplies, small prototypes or test circuits can live on my desk. More than that lives in a lab 100 to 300 feet away.

 

[f95toli]

So for an application that has an inherent portability requirement - for example, if I needed quantum digital processing to help identify or filter targets for a missile - would you say that we are already pretty close to a solution? Say within a decade?

No, not within a decade; perhaps never. It is not only about engineering but also if there is actually any reason to put a QC in a missile? The only class of circuits I know of which might be relevant would be in quantum machine learning; but I don't think you would need to solve a QML problem from start to end in a flying missile.
For the foreseeable future the way to access a QC will be via cloud access, very few institutions will actually buy the hardware. In part this is about business models, Google&co would much prefer if EVERYTHING was cloud access which was then access via very simple terminals, including applications that are now running on your desktop computer.

Also, you can see that in an application of this sort, you would still need to front end the quantum device with conventional electronics - but not necessarily a full computer processor?

The QPU is of course agnostic when it comes to what is sending/receiving signals. However, note that there are differences between how NISQ QPUs and error corrected QPUs operate, the former requires very fast communication and data transfer between the QPU and the supporting electronics (because of the way NISQ algorithms work) which is why we typically use a mix of FPGAs and regular CPUs

When I say "lab", I am referring to the alternative to my office. My computer, logic scope, small power supplies, small prototypes or test circuits can live on my desk. More than that lives in a lab 100 to 300 feet away

A better comparison would be a HPC in a data centre; you wouldn't typically need to work the same room as several racks worth of computer hardware
The model people have in mind for using quantum computers is something like existing ways to access AWS or Azure; you already have GPUs .TPUs, FPGAs etc and you can access these as needed when a specific platform will speed up the execution of a specific part of your application (TPUs for ML, FPGAs for DSP etc).
The QPU will "only" be another platform in this list.

 

[.Scott]

No, not within a decade; perhaps never. It is not only about engineering but also if there is actually any reason to put a QC in a missile?

I was around 50 years ago when people were saying the same kind of things about regular computers. At that time, computers lived in "Data Centers", usually position on the ground floor in rooms with walls made of glass.

Although the missile may not require QM for target recognition, it might have other potential applications for it. For example, if the missile knows that the target has the potential of being defended in many ways and that missiles approaching it are targeted for destruction or misdirection, it may need circuitry to create potential attack strategies applicable to the specifics of the situation. If there's a trillion, trillion potential paths to completing the mission, perhaps Grover's Algorithm can be used to identify a selection of promising ones for further classical evaluation.

For the foreseeable future the way to access a QC will be via cloud access, very few institutions will actually buy the hardware. In part this is about business models, Google&co would much prefer if EVERYTHING was cloud access which was then access via very simple terminals, including applications that are now running on your desktop computer.

I'll go with "near future". No doubt, the NSA already has a copy of anything promising that existed as recently as a few years ago - and no doubt it will maintain that status.

As far as Google is concerned, they have done quite well in adapting their product to many major markets. A couple of years ago, I remember they even introduced something that could hold some US classified information. And certainly, these recent ransomware attacks provide an incentive for people to hand their data management over to more expert hands.

But I believe there will always be demand in both the consumer and business market places for personal computing and data storage.

This is what is foreseeable:
1) Initially, very few businesses will purchase their own QM processors because:
1a) They're clunky machines with bizarre maintenance requirements; and
1b) They don't support a whole lot of business applications.
2) The equivalent of a "killer app" will be found for these QM devices. Then lot's a businesses will want access to these machines - and they will go to the cloud.
3) A market will exists to provide dedicated QM devices to on-site labs (or "data centers"). And that market will be served.

 

[f95toli]

I was around 50 years ago when people were saying the same kind of things about regular computers. At that time, computers lived in "Data Centers", usually position on the ground floor in rooms with walls made of glass.

Although the missile may not require QM for target recognition, it might have other potential applications for it. For example, if the missile knows that the target has the potential of being defended in many ways and that missiles approaching it are targeted for destruction or misdirection, it may need circuitry to create potential attack strategies applicable to the specifics of the situation. If there's a trillion, trillion potential paths to completing the mission, perhaps Grover's Algorithm can be used to identify a selection of promising ones for further classical evaluation.

Possibly, but as far as I am aware there are no known QC algorithms that would be useful for this. My understanding of quantum machine learning is that the QC are (potentially) very good for training, they are not necessarily faster at actually executing the algorithm .
It is important to understand that quantum computers are currently very, very specialised machines that are only better than a conventional computer for certain very specialised tasks. Now, some of these task include very important scientific problems (in e.g. quantum chemistry) as well as certain classes of optimisation problems which are relevant for e.g. finance and logistics (which is driving a lot of investment in this).
99% of the time a conventional CPU is going to be better and faster and that is unlikely to change unless we discover new algorithms that are more applicable to "general" problems.

A last point I'd like to make is that the engineering challenges involved in shrinking a QC down to "desktop" size are mostly NOT about the quantum processor itself (which is not very big), but about the supporting electronics/infrastructure. All implementations will require hundreds or thousands of signals in/out of the processor and these are either in the microwave regime or optical signals generated by lasers (which need to be very good, you can't use a laser diode).
All of this exists but is bulky (and expensive) despite many years of commercial pressure to make cheaper and more integrated electronics.
Right now, if you buy state-of-the art custom equipment you still need several racks worth of kit to control say a hundred qubits.

 

[berkeman]

Thread closed temporarily for Moderation of the latest post...

 

[berkeman]

After some spam cleanup, this particular thread will remain closed (not the fault of any posts still visible in the thread). Please start a new QC thread if you want to discuss technical QC details. Thanks.