Three peer-reviewed experiments published during 2026 are refining the physical picture behind the Landauer limit, the minimum energy associated with reliably erasing one bit under standard equilibrium assumptions. The results range from silicon memory cells that cannot reach the ideal limit to superconducting vortex memory and a feedback-controlled device with an apparent sub-limit measurement. None demonstrates a Bitcoin mining ASIC operating at a new J/TH record.
Background video: theoretical physicist and philosopher Wayne Myrvold discusses entropy, Maxwell’s demon and Landauer erasure in a 2025 interview. The discussion begins around 1:10:24; the recording is not a demonstration of the 2026 experiments.
The basic number is only the starting point
For a one-bit memory initially in thermal equilibrium, the familiar bound is E ≥ kBT ln 2, where T is absolute temperature. At about 300 kelvin, it is approximately 2.87 × 10−21 joule per bit erased. At 4.2 kelvin, the ideal expression falls to roughly 4.02 × 10−23 joule. Both figures refer to an idealized information-erasure operation—not an entire transistor, Bitcoin hash or air-cooled mining facility.
1. Silicon DRAM reveals a practical barrier
In Physical Review Letters on March 20, Takase Shimizu and colleagues at NTT measured erasure using silicon DRAM capacitors with single-electron charge-counting capability. Slower erasure alone did not bring their cells to the Landauer bound. The researchers traced the gap to an inability to prepare the cell’s initial state in thermal equilibrium, which prevented a truly quasistatic protocol. Lower tolerated error probabilities also worsened measured efficiency.
The finding matters because practical hardware must preserve reliable states, not merely minimize an ideal algebraic expression. A design that looks thermodynamically optimal on paper may still face stricter device-level restrictions.
2. A superconducting vortex approaches the ideal in modeling
A Nature Communications study published September 24 investigated a memory bit encoded by a vortex in a nanoscale double-well potential. Researchers manipulated it at 4.2 K and analyzed the work needed to change its state. Their numerical results placed an erasure-like operation within an order of magnitude of Landauer’s energy scale.
The distinction is important: the experiment used a steady sinusoidal drive incompatible with a formal Landauer erasure protocol. Its near-limit erasure estimate came from modeling, not a measured full reset cycle in a scalable commercial memory array. Refrigeration, control electronics, speed and state retention would have to be included in any computing-system efficiency claim.
3. Feedback shifts the apparent threshold
A separate Physical Review Letters paper published September 30 studied a feedback-driven microresonator used as a one-bit memory. By introducing hysteretic bias, researchers reported reset energy more than 20% below kBT0 ln 2, where T0 is the physical bath temperature.
The result does not overturn thermodynamics. The memory was maintained in a nonequilibrium environment with an embedded feedback mechanism, making the standard equilibrium comparison incomplete. A complete energy and entropy budget must consider the information engine and its surroundings.
What the three studies mean for Bitcoin ASIC efficiency
A Bitcoin miner performs repeated Double SHA-256 calculations across silicon logic, memory, power delivery and control electronics. Its industry metric is joules per terahash at a specified measurement boundary. Landauer’s per-bit-erasure scale is not an ASIC J/TH rating, and moving a laboratory memory to liquid-helium temperatures does not erase the cost of refrigeration.
The studies help define mechanisms and boundaries for long-run Bitcoin mining thermodynamics research, but they do not move the baseline in our 1 J/TH scenario analysis. Compare their memory physics with the AQFP SHA-256 simulation and the new 3.27× Double SHA-256 architecture study: the latter two at least model hashing hardware directly.
The path to a defensible PEI revision remains measured, sustained Double SHA-256 work, reported in comparable chip-, miner- and site-level J/TH units. Until such measurements appear, the new experiments are fundamental research findings, not mining-hardware efficiency breakthroughs.
Primary research: Shimizu et al., Physical Review Letters 136, 117103 (March 20, 2026); Fridman et al., Nature Communications (September 24, 2026); “Shifting Erasure Cost below the Landauer Bound with a Demon-Biased Thermal Bath,” Physical Review Letters 137, 147101 (September 30, 2026).

Leave a Reply