Why Quantum Computing Won’t Replace Classical CPUs—and What It Can Actually Solve

Mike Mayberry, Intel senior vice president, CTO and director of Intel Labs, oversees the company’s frontier technology investments and research direction.

Quantum Computing Is Not a Panacea

Mike states that quantum computing will not make existing classical CPUs obsolete; traditional computing remains essential for deep learning and future AI.

However, quantum computers excel at problems that are currently intractable, such as material simulation (e.g., catalysts, new drugs, room‑temperature superconductors) and certain mathematical challenges like post‑quantum cryptography.

Large‑Scale Commercialization Needs Ten Years

According to Mike, widespread commercial use of quantum computers is still about ten years away.

In the interim, quantum computers can tackle niche, hard‑to‑solve problems that could dramatically impact the world, such as developing catalysts to alter fuel composition or capture CO₂, which would address climate change.

Intel focuses not on the sheer number of qubits but on error‑correction capabilities and qubit stability.

Key Technical Challenges

Achieving reliable error correction to extend qubit lifetimes for meaningful algorithms.

Implementing local qubit control without long cables.

Designing efficient routing and placement of qubits within a physical system.

Connecting thousands of qubits into a scalable architecture.

Why Intel Targets Material Simulation

Materials are inherently quantum systems, making quantum simulation a natural research entry point; Intel has pursued this for over a decade.

Quantum Chip Progress

Intel’s 49‑qubit test chip launched in January is currently undergoing feature testing, with a second version planned for later this year.

Evaluating Quantum Chip Leadership

Beyond qubit count, the overall chip performance is judged by metrics such as qubit initialization, operational lifetime, and the ability to entangle and execute algorithms before decoherence.

Probabilistic Computing and Bayesian Networks

Mike notes that Bayesian networks are essentially probabilistic models rooted in statistics; advancing beyond them requires handling uncertainty and integrating common‑sense world knowledge.

Safety in Autonomous Driving

To improve safety, Intel suggests embedding fail‑safe mechanisms like automatic braking and attention‑preserving configurations in autonomous systems.

R&D Investment and ROI

Research projects are evaluated and a subset (about 25%) is selected for development; early validation helps identify low‑ROI projects before large investments.

Future AI Aspirations

Mike is interested in reducing the data required for machines to achieve confidence comparable to humans, aiming to narrow the gap between human and machine learning efficiency.

—

Source: NetEase Smart (公众号 smartman163), Issue 77, April 2018