The most surprising thing about quantum computing is that farmland, despite not proving itself particularly useful, has already given rise to several startups interested in building something other than qubits. It may be easy to dismiss this as opportunism – the pursuit of money in a publicized situation quantum computing. However, it may be helpful to take a look at the issues these startups are focusing on, as they may be indicative of dry issues in quantum computing that have not yet been solved through one of these. Most of the heavyweight companies are interested in that field—companies like Amazon, Google, IBM, or Intel.
In the case of a UK-based company called Riverlane, the unresolved piece that is being addressed is the copious amounts of classical calculations that are being required to produce quantum hardware images. In particular, it focuses on the immense amounts of information processing that will be required for a key part of quantum error correction: detecting when an error has occurred.
error detection vs information
All qubits are fragile, losing their environment either during operation or as they age. It doesn’t matter what the era is – cold atoms, superconducting transmon, it doesn’t matter – those error charges put a dry restriction on the amount of calculations that can be done before the error becomes inevitable. It controls almost every useful computation running on current {hardware} qubits.
The most authoritative approach to this is work known as logical qubits. These involve linking together a few hardware qubits and spreading quantum data between them. Alternative hardware qubits are related so that they are able to observe errors affecting the information, allowing them to be corrected. It can drive dozens of hardware qubits to form a single logical qubit, meaning that even the most critical existing systems can support about 50 hard logical qubits.
Steve Brearley, Riverlane’s founder and CEO, told Ars that error correction not only strains the qubit hardware; It equally cleverly emphasizes the classical part of the components. To track the components, each measurement of the perturbed qubit must be processed to arrive at and interpret any faults. To do even one of the simplest interesting calculations we would need more or less 100 logical qubits, which means tracking thousands of hardware qubits. Performing additional microscopic calculations would probably be brutal for 1000 logical qubits.
Error-correction information (called syndrome information in the field) must be read between each operation, yielding a large amount of information. “At scale, we’re talking hundreds of terabytes per second,” Brierley said. “On a million physical qubits, we will process about a hundred terabytes per second, which is Netflix global streaming.”
Additionally it needs to be processed at real age, otherwise calculations are halted waiting for error correction to occur. To avoid this, mistakes should be detected at the actual age. For transmon-based qubits, syndromic information is generated more or less every microsecond, so the real age way ends up processing information – possibly terabytes of it – with a frequency of about a megahertz. And Riverlane was founded to serve hardware that is capable of handling it.
dealing with information
The components the company developed are described in a paper it posted on arXiv. It is designed to take care of syndrome data, then alternative hardware has already converted analog indicators into virtual mode. This allows Riverlane’s hardware to sit outside any low-temperature hardware that is required for some methods of physical qubits.
That information is administered through a set of rules, called a “collision clustering decoder” in the paper, that perform fault detection. To demonstrate its effectiveness, they implemented it analogous to a traditional Garden Programmable Gate Array from Xilinx, where it only occupies 5 percent of the chip but can take care of logical qubits built from about 900 hardware qubits (this Simulated on the case).
The company also demonstrated an optimized chip that processes a very large logical qubit, occupying just a tiny fraction of a square millimeter and consuming only 8 milliwatts of energy.
Any one of these variations is highly specialized; They simply feed fault data to alternative parts of the components to behave. So, this is a very targeted answer. However it is also highly versatile as it works with many error-correcting codes. Notably, it also integrates with techniques designed to monitor a qubit analogous to other physics, including cold atoms, trapped ions, and transmons.
Brierley noted, “I think it was a bit of a puzzle in the beginning.” “You’ve got all these different types of physics; how are we going to do this?” This has not been a big problem. “One of our engineers was working with superconducting qubits at Oxford, and in the afternoon he was working with iron trap qubits. He came back to Cambridge and he was very excited. He said, ‘They’re using the same controls. There are electronics.” It appears that, regardless of the physics interested in controlling the qubits, everyone borrowed the same hardware from a different farm (Brearley noted that it was Xilinx radiofrequency equipment built for 5G. The system-on-a-chip prototype deployed below makes it much easier to integrate Riverlane’s custom hardware in a number of ways.