PrimerHow Far Are Quantum Computers from Breaking RSA-2048?Part 3 of 3

What Challenges Remain in Building a Quantum Computer Capable of Breaking RSA 2048?

Summary

Breaking RSA-2048 is held back not by quantum theory but by the engineering of a million physical qubits with full error correction: scaling chips, cryogenics, microwave control and real-time decoding all have to advance together. Chiplets, inter-chip links and larger dilution refrigerators make these problems solvable with sustained investment; the industry expects such a machine in the 2030s, so classical public-key cryptography needs post-quantum replacements deployed ahead of time.

On this page
  1. Quantum Chip
  2. Refrigeration System
  3. Qubit Control System
  4. Error Correction System
  5. Quantum Computers Capable of Breaking RSA-2048 Will Likely Appear in the 2030s
  6. References

Today’s more mature superconducting quantum computers typically have on the order of hundreds of qubits. As noted in Part 1, quantum computers can accelerate integer factorization. Shor’s algorithm requires roughly 2N qubits to factor an N‑bit integer, plus ancilla registers and other overhead. To break RSA‑2048, the total would be on the order of several thousand qubits.

That might sound close to success—but not quite. To complete the computation, qubit coherence time must exceed the time needed for all required operations. Today, superconducting‑qubit coherence times are usually tens to hundreds of microseconds, while gate times are tens to hundreds of nanoseconds. Shor’s algorithm requires on the order of 10¹² quantum‑gate operations, far exceeding typical coherence times. Even if we could finish within coherence time, gate fidelity would still limit performance: at 99.9% per gate, tiny errors accumulate over 10¹² operations and ruin the result.

Therefore, we must rely on quantum error correction to break through these physical bottlenecks. Taking surface code as an example, by encoding a large number of physical qubits into one logical qubit, it effectively extends “usable coherence time” and offsets cumulative errors in gate operations. Based on current error rates and error correction schemes, one logical qubit typically requires thousands of physical qubits. In other words, to obtain thousands of logical qubits to run Shor’s algorithm, the total number of physical qubits actually required will reach the million level.

Next, we will introduce in detail the technical obstacles that each core component of quantum computers needs to cross to expand the number of superconducting qubits from hundreds to millions.

Quantum Chip

The quantum chip—where qubits (Josephson junctions) reside—is also called a quantum processor. Scaling to more qubits faces three main difficulties: wiring, crosstalk, and semiconductor yield.

Wiring Problem

Each qubit requires multiple lines (control and readout), and couplers interconnect qubits. On a two‑dimensional chip, as qubit count rises, wiring complexity grows nonlinearly. In highly connected layouts, control lines for central qubits must snake around peripheral ones, dramatically increasing chip area.

Wiring complexity grows nonlinearly with qubit count. Figure: Wiring complexity grows nonlinearly with qubit count.

Crosstalk Problem

Crosstalk—mutual interference between qubits—causes decoherence and grows nonlinearly with system size. Common forms include:

  • Classical crosstalk: control frequencies are too close, causing unintentional drive. (In quantum processors, each qubit is driven at a specific microwave frequency.)
  • Quantum crosstalk: residual coupling that fails to turn fully off (analogous to current leaking through an “open” classical switch).
  • Global crosstalk: interference from external physical processes such as cosmic rays or phonons.

To mitigate crosstalk, designers may add isolation regions and shielding, and improve coupler devices to achieve better off‑state isolation. Enhancing the control/readout stack—especially frequency planning—also helps reduce crosstalk during parallel two‑qubit gates.

Crosstalk grows nonlinearly with qubit count and harms coherence. Figure: Crosstalk grows nonlinearly with qubit count and harms coherence.

Device Yield Problem

If looking only at area, the area of quantum chips should be linearly related to the number of qubits. However, due to the need to handle wiring and crosstalk problems, actual chip area often approaches square growth with the number of qubits. In other words, the more qubits, the more nonlinearly the chip area expands.

More troublesome is that quantum bits are extremely sensitive to defects, and even a 1% failure rate will make the entire system unusable. If there are defects inside or on the surface of the chip, they may couple with quantum bits, reducing their coherence time. In the field of micro-nano processing, there is a basic rule: the larger the chip area, the lower the yield, and the manufacturing difficulty of large-area chips increases exponentially. For superconducting quantum chips, although their manufacturing process can borrow mature equipment and process flows from the semiconductor industry, the extreme sensitivity of quantum bits to manufacturing defects makes the yield problem a huge challenge.

Area scales ~N² as qubit count grows; yield typically falls with die area (illustrative values). Figure: Area scales ~N² as qubit count grows; yield typically falls with die area (illustrative values).

Hope on the Horizon: Modular Design and Inter-Chip Interconnection

Wiring, crosstalk, and yield all worsen nonlinearly with scale. Building a million qubits on a single chip is therefore unrealistic.

A more feasible path is modular: first build chiplets with thousands of physical qubits (enough to form reliable logical qubits), then connect them with inter‑chip interconnects. This reduces the single‑die challenge from “hundreds to millions” down to “hundreds to thousands.”

However, this idea also brings new problems. Quantum bits are very fragile and must work in low-temperature environments around 10 millikelvin. If each chiplet is placed in an independent dilution refrigerator, then to achieve interconnection between chiplets, signal lines would need to be led out from the low-temperature environment of one refrigerator to room temperature, then enter the low-temperature environment of another refrigerator.

This “low temperature ↔ room temperature ↔ low temperature” signal transmission path will introduce large thermal loads and noise, thereby destroying the state of qubits. If all chiplets are placed in the same dilution refrigerator, then we need an extremely powerful dilution refrigerator to accommodate thousands of chiplets, and manufacturing such a large‑scale dilution refrigerator is itself a new challenge.

Chiplet links across separate refrigerators introduce thermal noise. Topology shown is illustrative. Figure: Chiplet links across separate refrigerators introduce thermal noise. Topology shown is illustrative.

Multiple chiplets in a single refrigerator with direct low‑temperature links. This centralized design needs ultra‑high‑power refrigeration. Topology shown is illustrative. Figure: Multiple chiplets in a single refrigerator with direct low‑temperature links. This centralized design needs ultra‑high‑power refrigeration. Topology shown is illustrative.

Going forward, either cross‑refrigerator links must suppress added noise, or dilution refrigerators must scale dramatically. Given today’s science and engineering, the latter—higher‑power, larger‑volume dilution refrigerators—appears more achievable.

Summary

If we adopt chiplets plus inter‑chip interconnects, the chip‑level challenge is expanding a single die from hundreds to thousands of qubits.

The good news: semiconductor manufacturing is a well‑developed technology stack, and related processes continue to improve. For example, advanced‑packaging 3D stacking can boost wiring density and interconnect capability. Better superconducting materials and processes, multiplexed designs, and smarter chip architectures (e.g., higher‑isolation couplers, more effective frequency planning) will also help.

Thus, expanding a single chip from hundreds to thousands of physical qubits, while challenging, is primarily an engineering bottleneck and looks relatively tractable. Currently, IBM has demonstrated a single chip with 1,000 physical qubits; large die size, however, raises yield and within‑chip reliability concerns for mass production.

Refrigeration System

Since the energy level difference of superconducting qubits is very small, quantum states are extremely susceptible to disturbances from the external environment and decohere. Currently known main interference factors include:

  • Passive thermal sources: Although qubits are in an environment close to absolute zero, they must be connected to room‑temperature electronic equipment through wires. Wires act like a “thermal bridge,” bringing heat from high‑temperature environments to low‑temperature areas.
  • Active thermal sources: Manipulating qubits requires sending microwave pulses, which are transmitted along wires. During transmission and attenuation, energy always converts to heat, which accumulates and heats the environment.
  • Thermal radiation: Even if wires and materials are isolated, there is still electromagnetic radiation coupling between quantum chips and the outside world. High-temperature environments will “radiate heat” to low-temperature chips.

Three main thermal interference sources for qubits. Figure: Three main thermal interference sources for qubits.

To isolate these effects as much as possible, current practice is:

  • Use materials with appropriate thermal and electrical conductivity to make wires;
  • Add filters and attenuators in signal paths to weaken unnecessary frequency bands and dissipated energy from control signal transmission;
  • Use dilution refrigerators for staged cooling (not directly cooling from room temperature to 10 mK, but sequentially through room temperature → 50 K → 4 K → 1 K → 100 mK → 10 mK), progressively shielding thermal sources.

Staged cooling in a dilution refrigerator: thermal buffers isolate heat, filters remove unwanted bands, attenuators absorb excess energy, and shields block radiation. Figure: Staged cooling in a dilution refrigerator: thermal buffers isolate heat, filters remove unwanted bands, attenuators absorb excess energy, and shields block radiation.

This approach currently supports systems with on the order of hundreds of physical qubits. To reach millions (e.g., thousands of chiplets with thousands of qubits each), a scaling problem emerges: each qubit needs wires, attenuators, and filters, and these scale roughly linearly with qubit count. While a single wire’s thermal leak is small, a million such components overwhelm cooling capacity. Beyond more powerful refrigerators, we need aggressive multiplexing, cryogenic CMOS (cryo‑CMOS), and low‑temperature superconducting electronics to cut wiring and power.

In short, the cooling power and physical space of existing dilution refrigerators are seriously insufficient. To support a million-bit system, the refrigerator’s power needs to increase by at least a million times or more. But such ultra-high-power, ultra-large-volume dilution refrigerators do not yet exist.

Refrigeration must scale from hundreds to millions of qubits. Figure: Refrigeration must scale from hundreds to millions of qubits.

If adopting a “chiplet design” + “low‑temperature inter‑chip interconnect” scheme, then the refrigeration system is a major engineering hurdle that must be crossed, requiring enormous R&D investment (likely in the tens of billions of dollars range).

Qubit Control System

As noted earlier, each superconducting qubit is controlled with microwaves, and different qubits are assigned different drive frequencies. Scaling from hundreds to millions raises several challenges:

  • Frequency crowding: The microwave spectrum range is limited. As the number of bits increases, frequency intervals are forced to decrease. When adjacent frequencies are too close, control signals may interfere with each other, causing crosstalk problems.

Adjacent drive frequencies can interfere, causing crosstalk. Figure: Adjacent drive frequencies can interfere, causing crosstalk.

  • Increased precision requirements: Narrower frequency bands mean control signals must be more stable, otherwise they will “overflow” to adjacent bits. Requirements for frequency stability and phase noise become more stringent.

Narrower bands demand better frequency stability and lower phase noise; small drifts can spill into adjacent qubits. Figure: Narrower bands demand better frequency stability and lower phase noise; small drifts can spill into adjacent qubits.

  • Soaring control complexity: Each qubit requires independent pulse control (amplitude, phase, timing). If there are a million qubits, there must be a million independent control channels. Currently, the hardware cost per channel is approximately ¥100,000/channel, with the long‑term goal of reducing it to ¥1,000/channel, otherwise the cost will be unbearable.

A million qubits require a million independent control channels. Figure: A million qubits require a million independent control channels.

These problems are essentially engineering bottlenecks. They have already appeared in small‑scale systems, and complexity increases linearly as scale expands.

However, the challenges in this direction are generally considered relatively optimistic compared to other issues in the industry, mainly for the following reasons:

  • Pattern of frequency crowding: Experiments show that within a few dozen qubits (≈60), frequency planning becomes rapidly more complex, but with more qubits, frequency reuse and careful layout can keep complexity manageable. So “frequency crowding” is not necessarily a hard barrier.
  • Limited gate precision requirements: Quantum computing does not require infinite precision. As long as two‑qubit gate fidelity can stabilize around 99.99%, it is sufficient to support quantum error correction. Although this still requires very high noise control for the control system, it is achievable with existing semiconductor technology. The only issue is that high‑speed digital‑to‑analog converters (DACs) capable of achieving this precision are currently too expensive. Whether costs can be reduced through large‑scale manufacturing remains to be seen.
  • Potential for hardware cost optimization: Current control stacks rely on superconducting coaxial cables, adapters, attenuators, and filters—expensive and bulky. Borrowing flexible‑substrate processes from semiconductors to mass‑produce integrated cryogenic wiring on low‑cost materials could cut both cost and volume.

In summary, the control system is another engineering bottleneck that must be crossed: million‑qubit systems require a million independent control channels, with frequency allocation, signal precision, and cost reduction all being key difficulties. Although relatively more optimistic compared to other issues, it still requires long‑term engineering investment and technological iteration.

Error Correction System

To make qubit and gate fidelity meet the requirements for completing large‑scale computations like factoring RSA, we must rely on quantum error correction (QEC): using thousands of physical qubits to form one logical qubit.

The theoretical framework of quantum error correction is mature and can significantly improve coherence time and gate fidelity of logical qubits. For example, Google’s Willow project demonstrated more stable logical qubits through error correction.

In the Willow processor, a logical qubit is not directly represented by a single physical qubit but is jointly encoded by a two‑dimensional physical‑qubit array. The array contains two types of qubits:

  • Data qubits: Used to carry logical quantum states;
  • Ancilla qubits: Through periodic operations to detect whether errors have occurred between data qubits.

The measurement results of these ancilla qubits do not directly reveal the logical state itself but can reflect whether errors such as bit flips or phase flips have occurred in the system. Combined with decoding algorithms, the system can determine where errors occurred and make corrections. In Willow’s experiments, researchers verified this key characteristic on hardware for the first time: when the physical‑qubit error rate drops below the threshold, theoretically, as long as engineering allows, the effective coherence time of logical qubits can be continuously extended by increasing encoding scale, and the fidelity of logical gates can be continuously improved. In other words, the ultimate limitations on coherence time and gate fidelity mainly come from engineering resources, not fundamental physical laws.

Many noisy physical qubits can encode a stable logical qubit; larger codes extend coherence as needed. Figure: Many noisy physical qubits can encode a stable logical qubit; larger codes extend coherence as needed.

However, in engineering implementation, decoding algorithms must be real-time: if errors accumulate too quickly, they will exceed the recoverable range. As the number of bits increases, error correction overhead grows nonlinearly, imposing extremely high requirements on classical hardware computing power and latency. Fortunately, error correction computation itself can be highly parallel, so theoretically bottlenecks can be alleviated by increasing computing power (i.e., “throwing money at it”).

The challenge is not only at the physical level but also in software‑hardware integrated engineering: since qubits continuously generate random errors that cannot be known before operation, circuits must dynamically adjust subsequent operations based on error information detected on‑site. At the same time, the error correction system must also implement real‑time scheduling and calibration of logical qubits and complete parallel decoding under extremely low latency. These requirements stacked together make fault‑tolerant quantum computing not only a hardware challenge but also a system engineering project of enormous scale and extremely high complexity.

QEC is a demanding systems problem: detect errors, decode in parallel, and correct within microseconds; compute demands grow with scale. Figure: QEC is a demanding systems problem: detect errors, decode in parallel, and correct within microseconds; compute demands grow with scale.

In summary, the quantum error correction system is a threshold that must be crossed: it requires thousands of physical qubits to form one logical qubit and relies on real‑time, parallel error correction and dynamic scheduling, imposing extremely high requirements on software‑hardware integration and representing an extremely challenging system engineering project. However, the entire process has no scientific limitations that cannot be solved—it is more a challenge of system engineering.

Quantum Computers Capable of Breaking RSA-2048 Will Likely Appear in the 2030s

In summary, to manufacture a million‑qubit computer capable of breaking RSA‑2048, the scientific challenge lies in the unavoidability of decoherence and gate errors, but quantum error correction theory has provided a clear solution path; the real difficulty is at the engineering level, including cooling, control, wiring, energy consumption, and real‑time implementation of quantum error correction. As scale expands, these problems will show nonlinear amplification, especially crosstalk, error correction overhead, and energy consumption. However, after a certain scale, modular design and inter‑chip interconnection can make complexity enter a regionally linear phase. Overall, there are no scientific “dead ends,” but engineering challenges are extremely large, requiring long‑term accumulation and huge investment. Currently, the industry generally expects such quantum computers to appear in the 2030s with high probability, and we agree with this judgment.

References

  1. “Postquantum Cryptography: The Time to Prepare Is Now!” Gartner Research(2024)
  2. Joseph, David, et al. “Transitioning organizations to post-quantum cryptography.” Nature 605.7909 (2022): 237-243.
  3. Kjaergaard, Morten, et al. “Superconducting qubits: Current state of play.” Annual Review of Condensed Matter Physics 11.1 (2020): 369-395.
  4. Castelvecchi, Davide. “IBM releases first-ever 1,000-qubit quantum chip.” Nature 624.7991 (2023): 238-238.
  5. Fowler, Austin G., et al. “Surface codes: Towards practical large-scale quantum computation.” Physical Review A—Atomic, Molecular, and Optical Physics 86.3 (2012): 032324.
  6. “Quantum error correction below the surface code threshold.” Nature 638, no. 8052 (2025): 920-926.