In short: Quantum computing went from Feynman's 1981 argument that nature cannot be simulated efficiently on classical machines, through Deutsch's formal model (1985) and the algorithmic breakthroughs of Shor (1994) and Grover (1996), to today's noisy 50-1,000 qubit superconducting and trapped-ion devices that are only now crossing the error-correction threshold.
Interviewers love the history question because the answer reveals whether you understand why quantum computers exist. The story is not "someone built a faster chip". It is a chain of ideas: a physics problem, a model of computation, killer algorithms, an error-correction theory, and finally hardware engineering that borrows heavily from semiconductor fabrication. If you are new to the basics, start with what is quantum computing and come back.
1. Feynman and the simulation problem (1980-1981)
Paul Benioff described a quantum-mechanical Turing machine in 1980, but the spark most people cite is Richard Feynman's 1981 talk "Simulating Physics with Computers". His argument was a counting one. The state of n interacting spin-1/2 particles needs 2n complex amplitudes. At n = 50 that is about 1015 amplitudes, roughly 16 petabytes at 16 bytes each. A classical computer must store and update all of them; a quantum system simply is that state.
Key fact: Feynman did not propose a faster factoring machine. He proposed a machine built from quantum parts because simulating quantum chemistry and materials classically scales exponentially in system size.
That motivation is still the most credible near-term application: molecules, catalysts and battery materials, not breaking encryption.
2. Deutsch makes it a model of computation (1985-1992)
In 1985 David Deutsch defined a universal quantum computer and showed it could simulate any physical process, strengthening the Church-Turing thesis into a physical principle. In 1992 the Deutsch-Jozsa algorithm gave the first problem where a quantum machine needs one query where a deterministic classical one needs up to 2n-1+1. It was contrived, but it proved the speed-up was real and set the template: put the input in superposition, apply an oracle, and use interference so wrong answers cancel.
3. Shor and Grover: why the field became urgent (1994-1996)
Peter Shor's 1994 algorithm factors an n-bit integer in time polynomial in n, using the quantum Fourier transform to find the period of ax mod N. The best known classical method, the general number field sieve, is sub-exponential but still super-polynomial. Since RSA relies on factoring, funding agencies and governments suddenly cared. This is the origin of today's post-quantum cryptography migration (NIST standardised ML-KEM and ML-DSA in 2024).
Lov Grover's 1996 algorithm searches an unstructured space of N items in about (π/4)√N queries. A quadratic, not exponential, speed-up, and provably optimal. Practical consequence: symmetric keys need doubling, so AES-256 stays comfortable. Many candidates confuse the two, so state the difference clearly: Shor is exponential for a structured problem, Grover is quadratic for an unstructured one.
Key fact: Breaking RSA-2048 with Shor needs on the order of thousands of logical qubits, which at realistic error rates means roughly a million physical qubits. Today's chips have hundreds to about a thousand.
4. The error-correction rescue (1995-2000)
Early sceptics argued that decoherence would kill any large computation. The 1995-96 answers from Shor and Steane (quantum error-correcting codes) and the threshold theorem of Aharonov, Ben-Or, Knill, Laflamme and Zurek changed that: if each physical gate errs less than a threshold (about 1% for the surface code), you can make logical error rates arbitrarily small by adding more qubits. In 2000 David DiVincenzo listed five requirements for a real machine: scalable well-characterised qubits, initialisation, long coherence relative to gate time, a universal gate set, and qubit-specific measurement.
5. First hardware: NMR, ions and superconducting circuits
The 1990s and 2000s tried many platforms. Liquid-state NMR let IBM and Stanford factor 15 into 3 × 5 on seven spins in 2001, but it does not scale because signal falls exponentially with spins. Cirac and Zoller's 1995 proposal used trapped ions as qubits; Wineland's group at NIST demonstrated a controlled-NOT shortly after. Superconducting circuits followed: Nakamura's charge qubit in 1999, then the transmon (Yale, 2007), which tolerates charge noise and became the industrial standard.
| Property | Superconducting (transmon) | Trapped ion |
|---|---|---|
| Qubit | Josephson-junction LC circuit | Hyperfine or optical level of an ion (Yb+, Ca+, Ba+) |
| Two-qubit gate time | about 20-100 ns | about 10-200 µs |
| Two-qubit fidelity | about 99.5-99.9% | above 99.9% demonstrated |
| Coherence | about 0.1-1 ms | seconds to minutes |
| Connectivity | nearest-neighbour on a chip | all-to-all within a chain |
| Operating condition | about 10-15 mK dilution fridge | room-temperature chip, ultra-high vacuum, lasers |
| Main players | Google, IBM, Rigetti, IQM | Quantinuum, IonQ |
6. Scaling up: from supremacy to below threshold (2019-2024)
In 2019 Google's 53-qubit Sycamore sampled a random circuit in about 200 seconds, a task estimated at thousands of years classically. Classical tensor-network methods later shrank that gap substantially, so treat "supremacy" as a benchmark, not an application. IBM scaled chip size: Eagle (127 qubits, 2021), Osprey (433, 2022) and Condor (1,121, 2023), then pivoted toward quality and modular designs. In December 2024 Google's 105-qubit Willow showed logical error falling as the surface-code distance grew from 3 to 5 to 7, the first convincing below-threshold result. Trapped-ion firms reported two-qubit fidelities above 99.9%.
7. Where semiconductor engineers fit in
Superconducting qubits are made with aluminium Josephson junctions on silicon or sapphire wafers using the same lithography, etch and metal deposition steps as CMOS, so yield, process control and defect analysis matter. Control electronics are a bottleneck: each qubit needs microwave lines, and thousands of coax cables will not fit in a fridge. That drives cryo-CMOS research where controllers operate at 4 K, plus flip-chip and through-silicon-via packaging. Layout of resonators and couplers is done in tools such as Cadence Virtuoso or Ansys, with electromagnetic simulation for crosstalk. Specialist design teams, for example the analog and mixed-signal group at Avecas design services, work on exactly this kind of cryogenic interface problem. To go deeper into how the physical qubit differs from a transistor, read qubits versus classical bits.
Plan your interview answer around three honest points: simulation was the original goal, error correction is the real bottleneck, and no one has yet shown a commercially useful quantum advantage. For chip-design foundations that make this career pivot easier, ChipXpert's VLSI training covers the CMOS and analog fundamentals.
Key fact: The transmon trades some anharmonicity for noise immunity; its transition sits near 5 GHz, which is why control uses ordinary microwave electronics at room temperature.
Want to go from history to hands-on? Explore the quantum and VLSI tracks on the CourseTron courses page and build the circuit-level intuition employers test for.
Frequently asked questions
Who invented quantum computing?
No single person. Benioff (1980) and Feynman (1981) motivated it, and Deutsch (1985) defined the universal quantum computer. Shor and Grover then gave the algorithms that made it matter.
What is the difference between Shor's and Grover's algorithms?
Shor factors integers in polynomial time, an exponential advantage that threatens RSA. Grover gives only a quadratic speed-up for unstructured search, so it halves effective symmetric key strength rather than breaking it.
Why are superconducting qubits the most common today?
They have nanosecond gates and are fabricated with semiconductor lithography, so existing foundry skills transfer. Their drawbacks are shorter coherence and the need for millikelvin cooling.
Has quantum computing beaten classical computers yet?
Only on contrived benchmarks like random circuit sampling. No useful, commercially relevant task has been solved faster, and fault-tolerant machines are still being developed.