Researchers at National Cheng Kung University have designed a more power-efficient circuit for Double SHA-256, the cryptographic operation used in Bitcoin mining. Their IEEE Access paper, published July 8, 2026, reports 3.27 times higher throughput per milliwatt than a particular earlier architecture under a common 28-nanometer power-estimation test. The work is a semiconductor design study, not a measured commercial Bitcoin miner.
Background video: learnmeabitcoin explains SHA-256 message expansion and compression. Published in 2020; it does not demonstrate the 2026 research chip.
Where the 3.27× advantage comes from
The study, by Chia-Chou Chuang, Yu-Hsiang Wu, Ying-Sheng Huang, Da-Huei Lee and Pei-Yin Chen, joins message expansion with compression rather than optimizing those circuits in isolation. Its two-round unrolled pipeline cuts the processing cycles in throughput-critical stages from 64 to 32. Fixed message formats and padding also create opportunities for precomputation, smaller storage and simpler arithmetic.
The authors evaluated CMOS implementations at 28 nm, 90 nm and 180 nm. At 28 nm, the design reached a reported peak data throughput of 1,587.60 Gbps, while its power-efficiency estimate at a common 36 MHz comparison point was 17.08 Gbps/mW. The latter was 3.27 times the comparable result for the architecture attributed to Zhang and colleagues. These are two separate operating conditions and should not be combined into a single miner specification.
| CMOS process | Reported power efficiency | Improvement versus Zhang design |
|---|---|---|
| 28 nm | 17.08 Gbps/mW | 3.27× |
| 90 nm | 4.31 Gbps/mW | 2.93× |
| 180 nm | 0.96 Gbps/mW | 3.31× |
Why gigabits per second cannot be treated as terahashes per second
The paper expresses throughput in Gbps, whereas commercial miners are rated in TH/s and energy per terahash (J/TH). A bit-throughput figure is not automatically the number of complete Double SHA-256 mining hashes delivered each second. Converting it into a credible miner J/TH rating requires a clearly defined hash-work accounting method, the actual finished silicon power, and the machine’s power-delivery and control overhead.
A 3.27× improvement in a particular benchmark also does not mean the next miner will use 3.27× less electricity. On equal and transferable throughput-per-watt assumptions, its reciprocal energy-per-unit would be roughly 69.4% lower than the benchmark—but translating that result into a modern ASIC requires physical validation. Read our ASIC-versus-site efficiency explanation for the accounting boundary.
Implications for the Power Efficiency Index
The relevance is architectural: a carefully reorganized Boolean datapath could improve energy efficiency without changing Bitcoin’s consensus algorithm. The paper also provides Verilog RTL and testbenches through its cited research repository, creating a path for independent verification. No public production miner specifications or deployment benchmarks accompany the paper.
For the 1 J/TH mining-efficiency scenarios, the result is a candidate design mechanism—not a new 2026 commercial efficiency observation. It complements our coverage of simulated superconducting SHA-256 circuits and the decade-long miner efficiency trend.
The decisive next experiment would implement an equivalent Bitcoin hashing workload on fabricated hardware, measure delivered Double SHA-256 hashes and all relevant electrical loads, and compare it with a contemporary commercial ASIC under matched conditions. Until then, 3.27× belongs to the research-design benchmark, not a miner leaderboard.
Primary source: Chuang et al., “High-Efficiency Double SHA-256 Architecture via Expansion–Compression Co-Design,” IEEE Access 14 (2026), pp. 103185–103194, DOI 10.1109/ACCESS.2026.3710840.

Leave a Reply