Superconducting Logic

Superconducting Josephson junctions can be used in place of transistors as a switching circuit. Logical circuits made of these elements can be vastly more energy efficient and faster than conventional CMOS circuits. There are two main families: rapid single flux quantum (RSFQ), and adiabatic quantum flux parametrons (AQFP). RSFQ chips can use extremely high clock rates (~100GHz), but require a current proportional to the number of Josephson junctions that can quickly become limiting (though many variants exist which address this in different ways). AQFP chips are ridiculously energy efficient. Like, 24 k_bT of energy per junction. Remember that the Boltzmann constant is 1.38 times 10 to the minus 23 J/K. AQFP makes the Landauer limit relevant. Anyway, AQFP is also faster than CMOS, just not RSFQ fast.

Spintronics

Spintronic circuits use magnetics (literally, electron spins) to store information instead of voltages or currents. These circuits can also be vastly more energy efficient than conventional CMOS.

• Two-dimensional spintronics for low-power electronics
• Mutual control of coherent spin waves and magnetic domain walls in a magnonic device

Quantum Computing

Standard CMOS, superconducting logic elements, and spintronics all rely on quantum effects, but they implement regular Boolean logic. Bits are zero or one, never both (well, unless your computer is broken). Quantum computers are a different beast.

Quantum bits, or qubits, exist in a superposition of zero and one. Concretely, this means that when you measure the qubit, there’s some probability that you will see a zero, and some other probability that you will see a 1. In the language of quantum mechanics, each state has an associated complex number, called an amplitude. (Amplitudes are converted to probabilities via the Born rule when measurements are made.) When you have multiple qubits, quantum mechanics requires that you don’t just have amplitudes for zero and one for each qubit independently; you have amplitudes for every possible arrangement of zeros and ones for all qubits. Thus the number of amplitudes grows exponentially with the number of qubits. So with just 10 qubits, a quantum system encodes 2¹⁰ = 1024 numbers. With 100 qubits, a quantum system encodes ~10³⁰ numbers. 300 qubits gives you more numbers than there are atoms in the universe.

We can’t measure these amplitudes directly, so this isn’t equivalent to having that much data in RAM. But since amplitudes can interfere, you can arrange things so that all the paths to the correct answer interfere constructively, while all the wrong answers have paths that interfere destructively. Then when you make your measurement, you are overwhelmingly likely to see the correct answer. This choreography of interference patterns is what quantum algorithms are all about. Note that this isn’t equivalent to trying all possible solutions in parallel.