Second-Generation Quantum Computers

by Dave D'Rave

The first-generation of quantum computers is being built right now.  Google, Rigetti, and IBM are all building superconducting loop-type quantum computers.  All of them say that they will have 50-qubit machines by January, 2020.

A 50-qubit quantum computer will be faster than any existing supercomputer for certain problems.  More importantly, we can expect that the number of reliable qubits in a system will increase by 30 percent per year for the next 20 to 30 years.  The trend is therefore that more and more problems will fall into the category of "A Quantum Computer is the Best Tool for That Job."

First-Generation Technology

The single most striking thing about current quantum computers is that they are very expensive.  Superconducting loop quantum computers typically require refrigerators which cost one million dollars, on top of the cost of the actual quantum chips and the room-temperature equipment which interfaces the system to the outside world.  Retail prices are quoted at $10 to $25 million, if you can get permission to buy one of these things.

Equally important, the price of a quantum computer is not expected to fall.  While the price of an individual qubit may decline, the number of qubits per processor is likely to increase faster.

This creates a certain "back to the future" situation.  For the time being, quantum computers will operate like old-time mainframes, such that users will submit their job to be run by a scheduler.  It will be interesting to see how the new generation of hackers adjusts to the concept of "four hour turnaround time."  It is also interesting to see whether the lack of privacy when using shared quantum computers will motivate the development of cheaper equivalents.

First-Generation Algorithms

As the number of qubits increases, the type and scale of problems which fit onto the machine will increase.  For example, a 50-qubit quantum computer will be superior to a classical supercomputer for certain math problems, such as solving the four color map theorem.

A 128-qubit quantum computer will be able to solve the 16-step traveling salesman problem in less than a second.  A 320-qubit quantum computer can solve the 32-step traveling salesman problem, etc.

A prompt (less than one second) known-plaintext attack on DES requires approximately 8,000 qubits.  (DES is the Data Encryption Standard, which was important in the 1980s and early-1990s.)

A quantum computer algorithm which breaks AES-128 requires 20,000 qubits, and AES-256 requires 40,000 qubits.  (AES stands for Advanced Encryption Standard.  This is a family of algorithms, and is widely used at this time.)

At current trends, quantum computers 20 years from now will have a major national security impact.  The question is: how large will the economic impact of cryogenic quantum computers be?

Second-Generation Technology

There are many candidate technologies to replace superconducting flux loops in next-generation quantum computers.  Given the cost and reliability advantages of room-temperature operation, I do not see how anything which needs to be at superconducting temperatures is viable.

It looks like optical non-linear thin films are the most promising technology for the second-generation of quantum computers.  These will have to be combined with integrated optical waveguides, photonic crystals, and plasmonic devices to achieve scalable, mass producible quantum computers.  These technologies already exist, and integrating them into a quantum information processor is a near-term development program.

Second-Generation Algorithms

When 100,000 qubit processors become available, it will be feasible to build machines which can brute-force many problems which are time-consuming for current technology.  Twentieth-century crypto systems, image processing, and semiconductor material design are obvious examples.  Less obvious are problems in non-linear physics, quantum chemistry, and meta-materials.

Security Issues

Since all of the proposed quantum computers use a classical computer to interface with the external world, they are vulnerable to the usual sort of exploits.  It is very unlikely that these problems will go away, as long as people are involved in operating the machines.