科技的未来：量子计算与QPU的“台积电”

Future of Tech: Quantum Computing and the "TSMC" for QPU Quantum Processor Units (QPUs) are likely to become the next important co-processor in data centers, sitting alongside CPUs and GPUs. CPUs will remain theworkhorse for general-purpose computing, while GPUs dominate highly parallel workloadssuch as AI. QPUs, in turn, could become essential for a set of problems that are toocomplex or too costly for classical processors to solve efficiently. QPUs leverage quantummechanics to solve exponentially complex problems classical computers cannot. Theanalogy is QPUs can search a 100-million-page phone book all at once while CPUs go pageby page.For its application in AI, its role is limited in the short term to hybrid workflowssuch as quantum kernels, feature mapping, and selected optimization or simulation-led tasks, but in the LT, when large-scale, error-corrected quantum computers emerge,QPUs could help accelerate parts of model training, optimization and possibly generativemodeling. Beyond AI, the clearest applications are in drug discovery, materials science,chemistry, and financial optimization. QPUs can be built using two leading modalities, superconducting circuits andtrapped ions. SuperconductingQPUs, used by IBM and Google, rely on patterned Josephson junction circuits fabricated with semiconductor style processes, but the leadingplayers keep this work in house. Superconducting QPUs benefit from fast gate speeds andhigh qubit counts, but is highly sensitive to noise.Trapped ionQPUs use microfabricatedion trap chips with patterned electrodes, plus growing optical integration, while the qubitsthemselves are individual atoms controlled by lasers and electromagnetic fields. Thistechnology uses real (rather than simulated) and hence high quality qubits, leading to longercoherence times and extremely high gate fidelity, but complex laser/optical control systemsmay hinder scaling. Infineon is the leading foundry in trapped ions, supplying QPU to Quantinuumand IonQ.Its roadmap is to build the enabling hardware - the ion traps, integrated photonics, and control electronics, and partner with the pure-play companies most likelyto scale trapped-ion systems commercially. IonQ may move to vertical integration withthe acquisition of SkyWater, but that also leaves substantial opportunities to Infineon forSkyWater’s other customers. If Quantum computing takes off with trapped iron paradigm,Infineon could benefit substantially. We believe that IBM’s investments in quantum computing will yield dividends.IBM(IBM) has spent the past decade advancing its vision of large-scale fault-tolerant quantumcomputing with one of the most comprehensive roadmaps in the industry. On the hardwarefront, IBM develops superconducting qubit quantum processors, including the 127-qubitEagle, 133-qubit Heron r1, and 156-qubit Heron r2 and r3. These quantum processingunits (QPUs) are then integrated into quantum systems and data centers. On the softwareside, IBM offers Qiskit, the world’s leading open-source platform that efficiently translateshigh-level algorithms into instructions optimized for quantum devices. By combiningcutting-edge hardware with powerful software, IBM Quantum delivers a full-stack quantumcomputing solution aimed at making practical quantum applications a reality. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS IBM: we rate IBM Market-Perform, with a target price of $330.IFX: we rate Infineon Outperform, with a target price of €52.Cambricon: we rate Cambricon Outperform, with a target price of CNY 2000.Hygon: we rate Hygon Outperform, with a target price of CNY 280. DETAILS Quantum processor units (QPUs) will run side by side with CPUs and GPUs.In the past decade, we witnessed the rise ofGPUs, thanks to AI. GPUs does not replace CPUs; in fact, majority of the day to day work is still running on CPUs , e.g. checkingemail, browsing, working on Word or Excel spreadsheets. But GPUs are excellent in one thing - parallel matrix computations,making them the ideal processor for AI. The GPU has become an essential co-processor of CPU. Quantum processors (QPUs) have high potential to be the co-processor of the future. QPUs don’t replace CPUs or GPUs, asmost of the classical problems are still easier or cheaper to be executed on CPU or GPUs. However, QPUs are superior or theonly option for a subset of intractable problems, making them irreplaceable in the future (Exhibit 1, Exhibit 2, Exhibit 3). This report contains two main parts. In the first part, we discuss the way quantum computing works and its promising futureapplications. In the second part, we discuss how to make a QPU using today’s semiconductor process, and highlight themethods. EXHIBIT 1:CPUs remain best for sequential work, which means tasks that must be done step by step. GPUs remainbest for highly parallel work, which means the same operation repeated many times at once. A QPU, is more likelyto become a third type of accelerator for a narrower class of prob