IBM Heron: The Processor Rewriting Quantum Error Rate Standards
When IBM announced Heron r2 in late 2024, the headline was a two-qubit gate error rate of approximately 0.1 percent. That number does not sound dramatic until you place it in context. IBM's Eagle chip, released in 2021, represented a major milestone as the first processor to break 100 physical qubits. Heron r2 reduced the error rate by roughly five times compared to Eagle. In a field where every decimal place matters for fault tolerance, that is a significant engineering achievement.
This post examines what is actually inside the Heron processor, why its architecture choices produce lower error rates, and what the progression from Eagle to Heron to Flamingo means for the long-term threat to cryptographic security.
IBM Heron r2 uses fixed-frequency transmon qubits on a heavy-hex connectivity lattice, cooled to 15 millikelvin. Its 0.1% two-qubit gate error rate and approximately 300-microsecond T1 relaxation time represent the most capable superconducting qubit processor IBM has shipped. Lower error rates matter more than raw qubit counts for eventual fault-tolerant operation.
The Architecture: Fixed-Frequency Transmons on a Heavy-Hex Lattice
What Is a Transmon Qubit?
IBM's quantum processors use transmon qubits: superconducting circuits based on Josephson junctions, a thin insulating barrier between two superconducting electrodes. When cooled to near absolute zero, these circuits exhibit quantum behavior. The energy gap between the ground state and first excited state defines the qubit's operating frequency, typically in the 4-6 GHz range.
Transmons trade sensitivity to charge noise (which plagued earlier Cooper-pair box designs) for slight sensitivity to flux noise. The result is a qubit type that is manufacturable at scale using standard semiconductor fabrication processes and that has well-understood decoherence mechanisms. IBM has fabricated transmon processors for over a decade, which means the engineering knowledge base around yield, calibration, and packaging is mature.
Fixed-Frequency vs. Tunable Qubits
Heron uses fixed-frequency transmons rather than frequency-tunable ones. Tunable qubits let engineers control the qubit frequency dynamically, which gives more flexibility in gate implementation. But tunable elements add flux bias lines, which introduce additional noise channels. Fixed-frequency qubits have fewer control lines and are less susceptible to certain noise sources, contributing to Heron's improved coherence times.
The tradeoff is that fixed-frequency architectures require more careful initial frequency assignment. If neighboring qubits have frequencies that are too close, they collide and produce unwanted interactions. IBM's device characterization and fabrication processes have improved enough that this tradeoff now works in Heron's favor.
The Heavy-Hex Lattice
IBM introduced the heavy-hex connectivity pattern starting with the Falcon processor. In a heavy-hex layout, qubits sit at the vertices and edges of a hexagonal lattice, but not every possible connection is implemented. Each qubit connects to at most three neighbors rather than the four or more that a square lattice would allow.
The benefit is reduced crosstalk. Fewer couplings per qubit means fewer unwanted interactions between qubits that share physical space on the chip. The cost is that not every pair of qubits can interact directly. Running algorithms that need qubit pairs not physically connected requires swap operations, which add gate depth and accumulate errors.
For most practical circuits, the error reduction from lower crosstalk outweighs the overhead of additional swaps. IBM's benchmarks on Heron show this tradeoff resolving in favor of the heavy-hex layout at current error rates.
Heron's Key Specifications
| Processor | Year | Qubit Count | 2-Qubit Gate Error Rate | T1 Relaxation | Key Milestone |
|---|---|---|---|---|---|
| IBM Eagle | 2021 | 127 | ~0.5% | ~100-150 µs | First processor above 100 qubits |
| IBM Osprey | 2022 | 433 | ~0.3-0.4% | ~150-200 µs | Largest qubit count at launch |
| IBM Condor | 2023 | 1,121 | ~0.3% | ~200 µs | First 1,000+ qubit processor |
| IBM Heron r1 | Dec 2023 | 133 | ~0.15% | ~250 µs | 5x error improvement over Eagle |
| IBM Heron r2 | 2024 | 133 | ~0.1% | ~300 µs | Best-in-class superconducting error rate |
| IBM Flamingo | 2024 | 156 | ~0.1-0.15% | ~300 µs | First chip in 3-chip quantum interconnect |
T1 and T2: Why Coherence Times Matter
T1 is the relaxation time: the average duration before a qubit spontaneously decays from the excited state to the ground state. T2 is the dephasing time: the average duration before a qubit loses its phase coherence without necessarily decaying. Both limit how long a quantum circuit can run before errors accumulate beyond a useful threshold.
Heron r2's T1 of approximately 300 microseconds sounds short, but it represents meaningful progress. Each two-qubit gate on Heron takes roughly 50-100 nanoseconds. At 300 microseconds T1, a qubit can participate in thousands of gate operations before relaxation becomes likely. This enables deeper circuits than were practical on Eagle, where T1 was closer to 100-150 microseconds.
The connection to cryptographic attacks is direct. Shor's algorithm on RSA-2048 requires circuits with millions of logical operations. Each logical operation requires many physical gates. Even with error correction reducing the per-logical-operation error rate, the total circuit depth demands coherence across millions of gate cycles. Heron's longer coherence times are a necessary step on that path, but not yet sufficient.
Quick Win
When comparing quantum processors, look at two-qubit gate error rates and T1 coherence times together, not qubit count alone. IBM Condor has 1,121 qubits but higher error rates than Heron's 133. For fault-tolerant algorithms relevant to cryptography, Heron's lower error rate makes it more capable than Condor for circuits requiring high-fidelity gates, even at lower qubit count.
Cryogenic Packaging: The 15-Millikelvin Requirement
Superconducting qubits operate in dilution refrigerators that cool the processor to approximately 15 millikelvin (15 thousandths of a degree above absolute zero). This temperature is colder than outer space, which sits at roughly 2.7 Kelvin. The cooling is required because thermal energy at room temperature is orders of magnitude larger than the energy gap between qubit states. At 15 millikelvin, thermal fluctuations are suppressed enough that the quantum states can persist long enough to be useful.
IBM's dilution refrigerators are custom-built by companies including BlueFors and Oxford Instruments. The refrigerators are large, expensive pieces of equipment. Scaling to thousands of qubits requires either enormous refrigerators or new approaches to chip-level connectivity. IBM's Flamingo chip, which is designed to link multiple processors through cryogenic interconnects, is one approach to this scaling challenge.
IBM Flamingo: The 3-Chip Quantum Interconnect
IBM Flamingo, released in 2024, takes a different approach to scaling than simply adding more qubits to a single chip. Flamingo is a 156-qubit processor designed to operate as part of a three-chip interconnected system. The chips communicate through a cryogenic quantum link, allowing qubits on different chips to interact as if they were on the same chip.
This modular approach addresses a fundamental engineering constraint. Chip yield degrades as chip size increases. A 1,000-qubit monolithic chip is harder to fabricate with acceptable yield than three 330-qubit chips linked together. The interconnect introduces its own error overhead, but IBM's preliminary results suggest the interconnect fidelity is high enough to be useful for multi-chip circuit execution.
IBM's planned Kookaburra processor extends this approach to a 1,386-qubit system built from modular linked chips. Kookaburra is targeted for the mid-2020s. For the full IBM hardware and software roadmap, see our post on IBM's quantum roadmap through 2033.
Benchmarking: Quantum Volume, CLOPS, and Layer Fidelity
IBM uses three primary benchmarks to characterize processor performance, each measuring a different aspect of capability.
Quantum Volume (QV) measures the largest square random circuit a processor can execute with greater than two-thirds probability of success. It captures a combination of qubit count, connectivity, and gate fidelity in a single number. IBM Eagle had a QV of 128 when released. Heron's lower error rates push QV significantly higher, though IBM has shifted emphasis toward newer metrics that better capture practical performance.
CLOPS (Circuit Layer Operations Per Second) measures how fast a processor can execute parameterized circuit layers. This is relevant for variational algorithms and hybrid classical-quantum workloads, where speed of circuit execution determines total runtime. Heron's improved gate fidelity contributes to higher CLOPS by reducing the need for repetitive error mitigation runs.
Layer Fidelity is IBM's newest metric. It measures the probability that a full layer of two-qubit gates across the entire processor executes correctly. This is the most direct measure of hardware quality for deep circuit execution. Heron's 0.1% per-gate two-qubit error rate translates to a per-layer fidelity that scales with the number of gates per layer and circuit depth.
Why Error Rates Matter More Than Qubit Counts for Crypto Threats
The misconception that qubit count is the primary variable for cryptographic threat is common. The actual bottleneck is fault-tolerant logical qubit quality, not raw physical qubit count.
Researchers at Google and IBM have published resource estimates for running Shor's algorithm against RSA-2048. A 2022 paper from Craig Gidney and Martin Eker estimated approximately 4,000 error-corrected logical qubits and around 10 billion Toffoli gates. The number of physical qubits required per logical qubit depends directly on the physical error rate. At a two-qubit gate error rate of 0.1%, typical surface code implementations require roughly 1,000-2,000 physical qubits per logical qubit. That puts the total physical qubit requirement for a full RSA-2048 attack in the range of 4-8 million qubits.
This means improving error rates does two things simultaneously. It makes existing qubit counts more useful for shallower circuits. And it reduces the total physical qubit overhead needed for fault-tolerant operation on deeper circuits. Heron's 0.1% error rate is an important improvement. It is not yet close to the thresholds required for full fault tolerance at cryptographic scale.
For a full analysis of the qubit requirements and timelines for breaking Bitcoin and RSA, see our post on how many qubits it would take to break Bitcoin.
Quick Win
Use IBM's open-source Qiskit Runtime to benchmark your own circuits on Heron hardware through IBM Quantum Platform. Measuring layer fidelity on your specific circuit structure gives more actionable data than published spec sheets, because circuit topology and gate patterns interact with Heron's heavy-hex connectivity in ways that aggregate error rates do not capture.
The Quantum Error Correction Path Forward
Heron r2's 0.1% two-qubit gate error rate is a stepping stone, not a destination. The surface code, the leading error correction scheme for superconducting qubits, requires physical error rates below roughly 1% to achieve any error suppression at all. Heron clears this threshold. But achieving the error rates needed for efficient fault tolerance, where a single logical qubit requires tens rather than thousands of physical qubits, requires physical error rates in the range of 0.001-0.01%.
IBM's roadmap projects continued improvement in gate fidelity through the late 2020s. The Kookaburra era targets error rates that would make fault-tolerant operation practical at useful logical qubit counts. Whether those projections hold depends on fabrication improvements, new qubit designs, and advances in cryogenic control electronics.
For a deeper look at how quantum error correction works and why it is the central challenge for cryptographically relevant systems, see our post on quantum error correction explained.
How QuanChain Monitors Hardware Progress Like Heron
QuanChain's Quantum Oracle tracks real-world hardware benchmarks, including published error rates and logical qubit counts, across all major quantum computing platforms. When hardware metrics cross predefined threat thresholds, the network automatically upgrades its cryptographic parameters. No user action or hard fork required.
Explore QuanChain Technology