IBM revealed today detailed plans to build a quantum computer that corrects errors and has significantly greater computational capabilities than existing machines by 2020. It hopes to make this computer available via the cloud to users by 2029.
Starling is a proposed machine that will consist of a series of modules, with each module containing a set chip, and be housed in a new datacenter in Poughkeepsie. Jay Gambetta is the vice president of IBM’s quantum initiative. He says that they have already begun building the space. IBM claims Starling is a quantum leap forward. The company wants it to be the world’s first large-scale machine that implements error correction. IBM, along with other companies such as Amazon Web Services and QuEra of Boston, will have overcome the biggest technical challenge facing the industry. IBM, like the rest of industry, still has many years of work to do. But Gambetta believes it has the edge because it has the building blocks for error correction capabilities on a large-scale device. This means improvements to everything from chip packaging to algorithm development. “We’ve cracked code for quantum errors correction, and we’ve now moved from science into engineering,” he says.
Correcting mistakes in a quantum machine has been a challenge for engineers due to the unique way that the machines crunch the numbers. Quantum computers use qubits instead of binary bits 1, or 0, which encode information. IBM uses tiny superconducting circuits that are kept at absolute zero and interconnected on chips to create qubits. Other companies have created qubits from other materials, such as neutral atoms and ions.
Quantum computer errors can occur, for example when the hardware alters one qubit and accidentally changes a neighboring quantum that shouldn’t be included in the computation. These errors accumulate over time. Quantum computers are unable to perform complex algorithms, which are the source of their scientific and commercial value. For example, they cannot perform highly precise chemistry simulations that can be used to discover new materials or pharmaceutical drugs.
But error correction requires significant hardware overhead. Error correction algorithms encode information in a constellations of physical qubits referred to as a logical qubit. Google’s surface-code algorithm is effective at correcting mistakes, but it requires 100 qubits in order to store one logical qubit. AWS’s Ocelot uses an error correction scheme which is more efficient and requires nine physical qubits for each logical qubit stored in memory. (The overhead for qubits that perform computations to store data is higher.) IBM’s error-correction algorithm(also known as a low density parity check code) will allow for the use of 12 physical qubits to every logical qubit, a ratio similar to AWS.
A distinguishing feature of Starling’s design is its ability to diagnose errors in real time, also known as decoding. Decoding is the process of determining if a quantum computer signal corresponds to a mistake. IBM has developed a decoding algorithm that can be quickly implemented by a conventional chip called an FPGA. Neil Gillespie, founder of UK-based quantum computing company Riverlane, said that this work “bolsters the credibility” of IBM’s method of error correction.
Other error correction schemes and hardware design are still in the running. Gillespie says that it’s not yet clear which architecture will win.
IBM plans to make Starling a computer that can perform computational tasks far beyond the capabilities of traditional computers. Starling will have 200 qubits that will be built using chips from the company. It should be able perform 100 million logical calculations with accuracy, while existing quantum computers are only able to do this for a few thousands.
Gambetta claims that the system will demonstrate error-correction at a larger scale than any other system. Previous error correction demonstrations by Google and Amazon involved a single logical chip. Gambetta calls these “gadgets experiments,” saying that “They’re small in scale.”
It’s still unclear whether Starling can solve practical problems. Some experts believe that you need to perform a billion error corrected logical operations in order to execute a useful algorithm. Wolfgang Pfaff is a physicist from the University of Illinois Urbana-Champaign. He says that Starling is “an interesting stepping-stone system”. “But it is unlikely that this will create economic value.” (Pfaff who studies quantum computer hardware has received research funding by IBM but isn’t involved with Starling. Pfaff says that the timeline for Starling is feasible. He says the design is “based on experimental and engineering realities”. IBM has come up with a design that is “pretty compelling.” But building quantum computers is difficult, and it’s likely that IBM will experience delays due to unexpected technical complications. He says that this is the first time anyone has done it. IBM’s road-map calls for building smaller machines first before Starling. It plans to demonstrate this year that robustly stored error-corrected data can be stored in a chip named Loon. The company will build Kookaburra next year, a module which can store information as well as perform computations. By 2027, the company plans to connect two Kookaburra modules into a larger quantum computing system, Cockatoo. After demonstrating this successfully, the next stage is to scale up and to connect around 100 modules in order to create Starling.
This strategy, says Pfaff, reflects the industry’s recent embrace of “modularity” when scaling up quantum computers–networking multiple modules together to create a larger quantum computer rather than laying out qubits on a single chip, as researchers did in earlier designs. IBM is also looking past 2029. Blue Jay, the next model, will be built after Starling. (“I like birds,” says Gambetta.) Blue Jay will have 2000 logical qubits, and be able to perform a billion logical functions.
