# Escape Velocity: the full atlas Atlas version 65775c8 (2026-10-05). Data CC0-1.0; text CC-BY-4.0. Cite the atlas version. Source: https://github.com/scape-velocity/escape-velocity # Technologies ## AI for science (ai/ai-for-science) Status: proposed. Readiness: not assessed. AI systems that propose, test and refine scientific hypotheses, designs and experiments well enough that their results are confirmed in the laboratory at a rate that speeds up discovery. Scope: In scope: models and agents used to generate and rank candidates (molecules, proteins, materials, experiments) and to close the loop with laboratory or simulation feedback. Out of scope: the energy cost of running them, covered by ai/energy-efficient-inference; the specific discoveries they enable, covered by technologies such as biotech/de-novo-protein-design and health/antibiotics-for-resistant-bacteria. Requires: - ai/energy-efficient-inference: Screening and agentic loops run many model calls per experiment, so the energy and cost per token set how far discovery loops can scale. Required by: biotech/de-novo-protein-design, health/antibiotics-for-resistant-bacteria. ## Energy-efficient AI inference (ai/energy-efficient-inference) Status: mapped. Readiness: TRL 6 (6 of 9). Readiness note: Efficiency gains are demonstrated in production serving (a 33x reduction in energy per median Gemini Apps prompt over one year, company-reported), but the order-of-magnitude hardware gaps below are at laboratory or prototype level. Running trained AI models to answer queries at a small fraction of today's energy per generated token, across the whole stack from the chip to the cooling plant, so that wider use of AI does not need proportionally more electricity. Scope: In scope: the energy drawn to serve a trained language model, measured per token or per query, including accelerator, memory, interconnect, host and cooling overhead. Out of scope: the energy of training (not mapped yet); the logic devices, data movement and cooling that inference depends on, covered by computing/beyond-cmos-logic, computing/low-energy-data-movement and enablers/heat-removal; and what AI is used for, covered by ai/ai-for-science. Metrics: - energy-per-token: current 0.72 J, target 0.07 J, distance to target 1.0 orders of magnitude. Conditions: GPU energy over the inference window divided by output tokens, as the source defines it; strongly dependent on model size, batch size, context and output length. The current value is for a 1B-parameter dense model, the favorable end of the range. Target rationale: One order of magnitude below the current value under the same conditions. Oviedo et al. (Joule 2026) estimate 8 to 20 times line-of-sight energy reductions across models, serving systems and hardware; 10 times sits inside that range and needs no new device physics. Evidence: vellaisamy2026characterization, oviedo2026energy. Gaps: - Moving weights and cache costs more than the arithmetic [critical, engineering, layer system, active]: Each generated token streams the model weights and the key-value cache through memory, so memory traffic, not arithmetic, tends to bind the energy per token. The best published electrical die-to-die link measures 0.65 pJ/bit in a prototype, an interface-only figure. Closing the gap means fewer bytes moved (quantization, cache compression, sparse attention) and cheaper bytes. Blocked by: computing/low-energy-data-movement. Evidence: park2024pjbit. - Decoding is bound by memory bandwidth, not compute [high, engineering, layer device, active]: Small batches leave the arithmetic units idle while the memory system is saturated. The measured token energy falls from 7.46 to 0.72 J/token as output length grows from 10 to 512 tokens at batch 16, because fixed costs are amortized; batching gains shrink as context grows (6.31x at 512 tokens of context against 1.17x at 4K for 10 output tokens). Blocked by: computing/low-energy-data-movement. Evidence: vellaisamy2026characterization. - CMOS logic has a ceiling of about two hundred times today's efficiency [medium, fundamental-limit, layer principle, open]: Ho, Erdil and Besiroglu estimate a ceiling of 4.7e15 FP4 operations per joule for CMOS microprocessors, roughly two hundred times current microprocessors, from switching, interconnect capacitance and leakage. The atlas has no sourced current FLOP-per-joule value for deployed accelerators yet, so this gap has no metric of its own. Blocked by: computing/beyond-cmos-logic. Evidence: ho2023limits. - Delivered power and cooling bound token output [medium, engineering, layer deployment, active]: At deployment scale the binding constraint can move from peak compute to delivered data-center power, cooling capacity and PUE. Embedded microfluidic cooling supports about 1e3 W/cm2; accelerator power density keeps rising. Blocked by: enablers/heat-removal. Evidence: wei2026microfluidic. - Idle capacity and small batches waste energy [high, engineering, layer deployment, promising]: Production energy per prompt includes idle machine capacity and data-center overhead, not only active accelerator power. Google reports a 33x reduction in energy per median Gemini Apps text prompt over one year from software efficiency and clean-energy procurement (the abstract states the combined effect on energy and does not separate the two). Further gains depend on batching, routing and model choice. Evidence: elsworth2025measuring, oviedo2026energy. Impact: - Benefit: Across models, serving systems and hardware, efficiency gains in sight could cut the energy of AI inference 8 to 20 times. At 1 billion queries a day with 10% long queries, demand would fall from 1.7 GWh a day to 0.8 GWh a day with efficiency interventions. Who: Data centers that serve AI models, and the electricity grids that supply them. Class: extrapolation. Assumptions: A bottom-up model of production serving from token throughput, node power and overhead, for frontier-scale models (more than 200B parameters) on H100 nodes; not a measurement. Unlocked by the target for energy-per-token. UN SDG: 7, 13. Evidence: oviedo2026energy. Requires: - computing/beyond-cmos-logic: Arithmetic in the accelerator is bounded by the switching energy of CMOS logic, which is estimated to allow only about two hundred times more efficiency than current microprocessors. - computing/low-energy-data-movement: Token generation reads the model weights and the key-value cache from memory for every token, so memory and chip-to-chip traffic carries a large share of the energy. Need: Energy per bit moved between memory and processor at or below 0.1 pJ, so that streaming weights costs a few watts per terabyte per second. - enablers/heat-removal: Delivered power and cooling capacity bound how many accelerators fit in a rack and therefore the tokens produced per site. Required by: ai/ai-for-science, neurotech/high-bandwidth-bci. The Alan Machine: [Building Alan: energy per token](https://the-alan-machine.github.io/the-alan-machine/building-alan/energy-per-token/index.html) ## De novo protein design (biotech/de-novo-protein-design) Status: scoping. Readiness: not assessed. Designing a protein with a chosen structure or function from scratch, by computer, so that the first designs tested in the laboratory work as intended. Scope: In: computational design of new protein binders, enzymes and assemblies that do not exist in nature, and the experimental success rate of the designs. Out: directed evolution of natural proteins, and delivery of the proteins into the body (biotech/in-vivo-gene-delivery). Metrics: - experimental-hit-rate: current 10%, target 90%, distance to target 1.0 orders of magnitude. Conditions: De novo protein binders, experimental success rate across the targets tested by the method's developers. Target rationale: The goal of the one-design-one-binder approach named in the BindCraft paper: with a success rate of 90%, testing three designs gives more than a 99.9% chance of at least one working binder (1 minus 0.1 cubed), so no high-throughput screening is needed. Atlas reasoning, not an agency target. Evidence: pacesa2025one, pacesa2025one. Gaps: - Success rate varies widely between targets and between groups [high, scientific-unknown, layer device, active]: The developers of one pipeline report experimental success rates of 10 to 100% depending on the target. An earlier study found the overall design success rate low and raised it nearly 10-fold with deep-learning filtering. An independent group testing another method on six targets, five designs each, reported that most targets gave no working binder. Blocked by: ai/ai-for-science. Evidence: pacesa2025one, bennett2023improving, jiang2025rfdiffusion. Requires: - ai/ai-for-science: Design methods are deep-learning models trained on protein structures. - biotech/low-cost-dna-synthesis: Every designed protein must be encoded as synthetic DNA before it can be expressed and tested. Required by: biotech/in-vivo-gene-delivery. ## In vivo gene delivery (biotech/in-vivo-gene-delivery) Status: proposed. Readiness: not assessed. Getting a gene, an editor or an RNA into the cells of a living body that need it, and only those cells, safely and at a dose that works. Scope: In: viral vectors, lipid nanoparticles and engineered capsids that carry genetic cargo to target cells in a living body. Out: the choice of the cargo and the design of new proteins (biotech/de-novo-protein-design is an input, not part of this entry). Requires: - biotech/de-novo-protein-design: Designed binders and capsid proteins can redirect vectors to specific cell types. ## Low-cost DNA sequencing (biotech/low-cost-dna-sequencing) Status: scoping. Readiness: not assessed. Reading the sequence of a human genome, at about 30-fold coverage, for a cost low enough that sequencing is a routine step in research, screening and diagnosis. Scope: In: instruments, chemistry and workflows that read DNA, and the cost per human genome. Out: the interpretation of the sequence, and DNA writing (biotech/low-cost-dna-synthesis). Metrics: - sequencing-cost: current 90 USD, target 10 USD, distance to target 1.0 orders of magnitude. Conditions: One human genome at about 30-fold coverage; platform developer's figure per gigabase scaled to a genome. Target rationale: Atlas working target, not set by an agency: one order of magnitude below the current value, where the sequencing cost of a genome would be comparable to a routine laboratory assay. NHGRI publishes the cost history but sets no target. Evidence: almogy2022cost, wetterstrand2023dna. Gaps: - Reagent price is not the cost of a genome [medium, cost, layer deployment, open]: The per-gigabase price quoted by a platform developer leaves out what NHGRI counts as production cost: labor, instruments amortized over three years, informatics and data submission. The atlas has no independent figure for the full cost of a 30-fold human genome today, so the current value is a derived, developer-reported number. Evidence: wetterstrand2023dna, almogy2022cost. Required by: health/multi-cancer-early-detection. ## Low-cost DNA synthesis (biotech/low-cost-dna-synthesis) Status: proposed. Readiness: not assessed. Writing long, accurate DNA of a chosen sequence at a cost and speed that let designs be built and tested without waiting or rationing. Scope: In: chemical and enzymatic synthesis of oligonucleotides, genes and assembled constructs. Out: reading DNA (biotech/low-cost-dna-sequencing) and delivering DNA into cells (biotech/in-vivo-gene-delivery). Required by: biotech/de-novo-protein-design. ## Plant genome engineering (biotech/plant-genome-engineering) Status: proposed. Readiness: not assessed. Making precise, heritable changes to the genomes of crop plants quickly and across species, without the long tissue-culture steps that limit many crops today. Scope: In: editing and transformation methods for crop plants and the regeneration of whole plants from edited cells. Out: the choice of trait (for example food/nitrogen-fixing-cereals) and field regulation of the resulting crops. Required by: food/nitrogen-fixing-cereals. ## Direct air capture (climate/direct-air-capture) Status: proposed. Readiness: not assessed. Plants that remove CO2 from ambient air and store it durably, at a cost low enough to remove gigatonnes per year. Scope: In: capture of CO2 from air by sorbents or solvents, with transport and geological storage. Out: capture at point sources, biological and mineral removal, and the storage site itself (climate/geologic-co2-storage). Gaps: - Projected cost stays in the hundreds of USD per tonne [critical, cost, layer deployment, open]: Design and learning-curve estimates put the levelized cost of capture in the hundreds of USD per tonne: 94 to 232 USD/t for a designed aqueous-KOH plant (capture only, a design not an operating plant), and 341 to 374 USD/t of CO2 net removed with transport and storage at 1 Gt-CO2 per year of cumulative capacity. No source found for the cost of a plant operating today. Closing the gap needs cheaper sorbents, lower regeneration energy and cheaper storage. Blocked by: materials/low-cost-co2-sorbents, climate/geologic-co2-storage. Evidence: keith2018process, sievert2024considering. Requires: - materials/low-cost-co2-sorbents: The sorbent sets the capture capacity, the regeneration energy and a large part of the plant's cost. - climate/geologic-co2-storage: Removal counts only if the CO2 stays out of the air for centuries; the storage is also part of the cost per tonne. ## Geologic CO2 storage (climate/geologic-co2-storage) Status: proposed. Readiness: not assessed. Injection of CO2 into deep rock formations at gigatonne-per-year scale, with monitoring that shows it stays there. Scope: In: saline aquifers and depleted fields, including injectivity, capacity, monitoring and leakage risk. Out: mineralization in basalt as a capture route, and capture itself (climate/direct-air-capture). Required by: climate/direct-air-capture. ## Beyond-CMOS logic (computing/beyond-cmos-logic) Status: proposed. Readiness: not assessed. Logic that performs a computation at far less energy per operation than silicon CMOS can, by steeper switches, reversible or adiabatic operation, or superconducting and other devices, down toward the thermodynamic limit. Scope: In scope: devices and circuits for logic operations, measured by energy per operation at the device and with cooling and power delivery included. Out of scope: moving data between memory and processor, covered by computing/low-energy-data-movement; removing the heat, covered by enablers/heat-removal; quantum logic, covered by the quantum domain. Metrics: - energy-per-operation: current not recorded, limit 2.87 × 10⁻²¹ J. Conditions: Energy dissipated per logic operation at the device. No sourced current value yet; see the report. Required by: ai/energy-efficient-inference. ## Low-energy data movement (computing/low-energy-data-movement) Status: scoping. Readiness: TRL 4 (4 of 9). Readiness note: The best figure found is a 28 nm prototype die-to-die interface; no deployed system figure is sourced yet. Moving bits between memory and processor and between chips at a small fraction of today's energy per bit, by shorter and denser electrical links, in-package memory and optical interconnects. Scope: In scope: energy per bit of memory interfaces and chip-to-chip links, including the transceiver on both ends. Out of scope: the logic that computes on the data, covered by computing/beyond-cmos-logic; the photonic chips themselves, covered by enablers/photonic-integration. Metrics: - energy-per-bit: current 6.5 × 10⁻¹³ J, target 10⁻¹³ J, distance to target 0.8 orders of magnitude. Conditions: Interface energy of an electrical die-to-die link over a silicon interposer, transceiver only; a prototype, not a deployed memory system. Target rationale: Atlas-set reasoning, not an agency target: at 0.1 pJ/bit, an accelerator streaming 10 TB/s (8e13 bit/s, an assumed HBM-class rate) spends 8 W on the link, against 52 W at the current 0.65 pJ/bit; this brings interconnect energy below a few percent of a kilowatt-class accelerator. Evidence: park2024pjbit. Requires: - enablers/photonic-integration: Optical links promise low energy per bit over longer distances than electrical links, and need integrated photonic chips to be made in volume. Required by: ai/energy-efficient-inference. The Alan Machine: [Building Alan: data movement](https://the-alan-machine.github.io/the-alan-machine/building-alan/data-movement/index.html) ## Cryogenic control electronics (enablers/cryogenic-control-electronics) Status: proposed. Readiness: not assessed. Electronics that operate at cryogenic temperatures next to the qubits and generate control signals and read out results, replacing one line from room temperature per qubit. Scope: In scope: cryo-CMOS, superconducting digital logic and photonic links used to control and read out qubits in the cryostat. Out of scope: the qubits (quantum/fault-tolerant-quantum-computer) and the decoder algorithm (quantum/real-time-qec-decoder). Required by: quantum/fault-tolerant-quantum-computer, quantum/real-time-qec-decoder. ## Dilution refrigeration (enablers/dilution-refrigeration) Status: proposed. Readiness: not assessed. Cryostats that cool devices to a few millikelvin and remove the heat that wiring, amplifiers and control electronics add at that temperature. Scope: In scope: dilution refrigerators and their cooling power at the stages where qubits and cold electronics sit. Out of scope: the control electronics themselves (enablers/cryogenic-control-electronics) and cooling at room temperature (enablers/heat-removal). Metrics: - cooling-power: current 0.002 W. Conditions: Cryogen-free dilution refrigerator with four parallel dilution units, at 102.48 mK; the power at the 10-20 mK qubit stage is far lower. Evidence: guan2025development. Required by: quantum/fault-tolerant-quantum-computer. ## High-flux heat removal (enablers/heat-removal) Status: scoping. Readiness: TRL 4 (4 of 9). Readiness note: Embedded cooling is shown on thermal test vehicles and small GaN-on-diamond devices; a review warns that records are not transferable to package-compatible, large-area, deployed platforms. Cooling that removes the heat of dense chips and stacked dies at the device, at heat fluxes well above those of air and cold-plate cooling, in a form that fits in a package and can be made in volume. Scope: In scope: embedded and near-junction cooling (microfluidic channels, manifolds, jets, two-phase flow, diamond spreaders), measured by heat flux removed at a stated temperature rise. Out of scope: the data-center plant and heat rejection; low-temperature refrigeration, covered by enablers/dilution-refrigeration. Metrics: - heat-flux: current 10⁷ W m⁻², target 10⁸ W m⁻², distance to target 1.0 orders of magnitude. Conditions: Heat flux a cooler removes per unit area of the device surface. Heated area, coolant, temperature rise and pumping cost differ across sources and must be stated. Target rationale: Atlas-set reasoning: 10 kW/cm2 over large package-compatible areas. DARPA's near-junction program reached above 40 kW/cm2 (4e8 W/m2) on small GaN-on-diamond devices, so a quarter of that record on deployable, area-scaled hardware would give stacked logic a tenfold margin over today's microfluidic range. Evidence: wei2026microfluidic, barcohen2019embedded. Required by: ai/energy-efficient-inference. The Alan Machine: [Building Alan: heat removal](https://the-alan-machine.github.io/the-alan-machine/building-alan/heat-removal/index.html) ## High-repetition-rate high-energy lasers (enablers/high-repetition-rate-lasers) Status: proposed. Readiness: not assessed. Lasers of megajoule class that fire several times per second at high wall-plug efficiency and for billions of shots. Scope: In: laser drivers for inertial fusion and the diode-pumped solid-state, gas and fibre approaches to them. Out: continuous-wave industrial lasers and ultrashort-pulse lasers of low energy. Required by: energy/fusion-power. ## Photonic integration (enablers/photonic-integration) Status: proposed. Readiness: not assessed. Many optical components, such as waveguides, modulators, lasers and detectors, made together on one chip with low loss at a cost that allows volume. Scope: In scope: integrated photonic circuits, such as silicon nitride, silicon and III-V platforms, and their loss, integration of sources and detectors, and fabrication. Out of scope: the systems that use them, covered by quantum/quantum-interconnect and computing/low-energy-data-movement. Metrics: - waveguide-loss: current 1.77 dB m⁻¹. Conditions: Silicon nitride waveguides with an 80 nm core, made in an anneal-free process at no more than 250 C. Evidence: bose2024anneal. Gaps: - Low loss and active devices in one process [medium, manufacturing, layer manufacturing, active]: Silicon nitride gives the lowest waveguide loss, and integrating III-V and silicon materials has made large-scale nitride circuits with lasers and detectors possible. Ultra-low loss has been reached in an anneal-free process at no more than 250 C, which is compatible with CMOS back-end steps; keeping that loss while lasers, modulators and detectors share the chip, at volume, is the open problem. Evidence: bose2024anneal, xiang2022silicon. Required by: computing/low-energy-data-movement, quantum/quantum-interconnect. ## High-power electronics (enablers/power-electronics) Status: proposed. Readiness: not assessed. Converters and switches that move megawatts to gigawatts at high efficiency and power density, from the grid to magnets, lasers and data centres. Scope: In: wide-bandgap (silicon carbide, gallium nitride) devices, converters and their cooling. Out: signal-level electronics and the logic of computing (computing/beyond-cmos-logic). Required by: energy/fusion-power. ## Fusion power (energy/fusion-power) Status: scoping. Readiness: not assessed. A power plant that produces net electricity from nuclear fusion, which first needs a fusion fuel target or plasma that releases several times more energy than is delivered to it. Scope: In: deuterium-tritium fusion plants, whether by inertial confinement (laser-driven targets) or by magnetic confinement (tokamaks and stellarators). Out: fission and fission-fusion hybrids. The fuel supply is energy/tritium-breeding; the magnets are materials/low-cost-hts-conductor. Metrics: - scientific-gain: current 1.5, target 100, distance to target 1.8 orders of magnitude. Conditions: Indirect-drive inertial confinement, laser energy delivered to the target, scientific breakeven. Target rationale: The high-gain requirement for an inertial fusion energy plant: a ratio of neutron yield to incident laser energy of about 100 (goncharov2025laser), stated independently as gains above 100 needed for a laser-fusion power plant (mcgeoch2025development). Gain must cover the driver's wall-plug efficiency, the thermal-to-electric conversion and the power recirculated to the driver, and still leave most of the output for the grid. Evidence: abushawareb2024achievement, goncharov2025laser. Gaps: - Gain of about 100 at power-plant repetition rates [critical, scientific-unknown, layer principle, active]: The record target gain is 1.5, from a single shot at the National Ignition Facility. A plant needs a gain of about 100, which in turn needs a high fraction of the laser energy coupled to the target and the loss mechanisms from laser-plasma instabilities held down. Broadband lasers show promise against those instabilities, and simulations predict gains above 100 with less than 1 MJ of argon fluoride laser energy in direct drive, with no experiment yet at that gain. Blocked by: enablers/high-repetition-rate-lasers. Approach: Direct drive with broadband argon fluoride lasers. Evidence: abushawareb2024achievement, goncharov2025laser, mcgeoch2025development. Requires: - materials/low-cost-hts-conductor: Compact magnetic-confinement designs need fields above 18-20 T, for which high-temperature superconductors are the enabling technology, and their cost and complexity are a large part of the reactor core. - energy/tritium-breeding: Deuterium-tritium plants must breed their own tritium; natural supply is negligible. - enablers/high-repetition-rate-lasers: A laser-driven inertial plant needs a driver that fires several times per second, not a few shots per day. - enablers/power-electronics: Magnet power supplies, pulsed power and the grid connection of a plant rely on high-power converters. ## Long-duration energy storage (energy/long-duration-storage) Status: proposed. Readiness: not assessed. Storage that holds grid energy for many hours to days or longer at a cost per kWh low enough to firm a grid supplied mostly by wind and solar. Scope: In: electrochemical (flow, metal-air), thermal, mechanical and chemical (hydrogen) storage with a duration beyond about 10 hours. Out: short-duration lithium-ion batteries for hours-scale shifting. ## Perovskite-silicon tandem photovoltaics (energy/perovskite-silicon-tandem-pv) Status: proposed. Readiness: not assessed. Solar modules that stack a perovskite cell on a silicon cell to exceed the efficiency limit of silicon alone, and keep that efficiency for the 25 to 30 year life of a solar module. Scope: In: monolithic and mechanically stacked perovskite-on-silicon cells and modules, and their durability. Out: single-junction silicon and all-perovskite tandems. ## Tritium breeding (energy/tritium-breeding) Status: proposed. Readiness: not assessed. A fusion plant that makes at least as much tritium in its own blanket as it burns, and recovers and recycles the fuel fast enough to keep its startup inventory small. Scope: In: breeding blankets, tritium extraction and the plant's fuel cycle. Out: the plasma or target physics itself (energy/fusion-power) and lithium-isotope enrichment supply, which is not yet in the atlas. Required by: energy/fusion-power. ## Nitrogen-fixing cereals (food/nitrogen-fixing-cereals) Status: scoping. Readiness: not assessed. Cereal crops that obtain most of their nitrogen from the air, through their own nitrogenase, an engineered symbiosis or a partnered microbiome, so that high yields do not depend on synthetic nitrogen fertilizer. Scope: In: maize, wheat, rice and other cereals that fix nitrogen in the plant or in a stable association with it. Out: fertilizer-applied microbial inoculants that supplement but do not replace fertilizer (they appear here only as an approach), and plant gene-editing tools, which are biotech/plant-genome-engineering. Metrics: - nitrogen-derived-from-atmosphere: current 82%, target 100%, distance to target 18 points. Conditions: Share of the crop's nitrogen from biological fixation of atmospheric nitrogen, measured with 15N methods. Target rationale: A crop that fully meets its nitrogen from the air is the 'N-self-fertilizing' crop described in the literature as capable of autonomous fixation, avoiding the need for chemical fertilizers. The target applies to elite cultivars at full yield, which is what the current value does not show. Evidence: vandeynze2018nitrogen, guo2022biological. Gaps: - Fixation shown in a landrace, not in high-yield cultivars [critical, scientific-unknown, layer system, active]: The high fixation share was measured in a landrace with aerial roots that secrete mucilage, grown in nitrogen-depleted soil. Moving the trait into elite cultivars at full yield, and under fertilization, has not been shown. Engineered root-associated bacteria that keep fixing in fertilized fields have been commercialized, but their abstract reports a yield gain over fertilizer alone, not a share of nitrogen from fixation. Approach: Engineered root-associated diazotroph that fixes nitrogen despite fertilizer. Evidence: vandeynze2018nitrogen, wen2021enabling. - A working nitrogenase inside plant cells [critical, engineering, layer device, active]: Expressing the nitrogenase components in plant mitochondria is the route to a crop that fixes its own nitrogen. Sixteen nitrogenase proteins have each been expressed and targeted to the mitochondrial matrix of a model plant, but the NifD component is the least abundant and a full working complex in a plant has not been shown. Blocked by: biotech/plant-genome-engineering. Approach: Fusion of NifD and NifK to equalize their abundance in plant mitochondria. Evidence: allen2017expression, guo2022biological. - Root-nodule symbiosis not yet engineered into non-legumes [high, scientific-unknown, layer principle, open]: Legumes host nitrogen-fixing rhizobia in root nodules. The objective of engineering nodulation in non-leguminous crops has not been achieved; the open questions are the signalling, infection and nodule-organogenesis programs. Blocked by: biotech/plant-genome-engineering. Evidence: huisman2019roadmap, guo2022biological. Requires: - biotech/plant-genome-engineering: Every route to a self-fertilizing cereal needs many coordinated edits to the plant genome. ## Antibiotics for resistant bacteria (health/antibiotics-for-resistant-bacteria) Status: proposed. Readiness: not assessed. New antibiotics, or other agents, that kill bacteria resistant to the drugs now in use, discovered and developed fast enough to keep pace with the spread of resistance. This entry maps research; it is not medical advice. Scope: In: small-molecule and peptide antibacterial discovery, including computational discovery, and agents against drug-resistant bacterial infections. Out: vaccines, bacteriophage therapy, and diagnostics. Computational discovery tools belong to ai/ai-for-science. Requires: - ai/ai-for-science: Machine-learning models screen and generate candidate molecules far faster than screening libraries by hand. ## Multi-cancer early detection (health/multi-cancer-early-detection) Status: scoping. Readiness: not assessed. A blood test that screens people without symptoms for many cancer types at once, finds cancers early enough to change outcomes, and rarely raises a false alarm. This entry maps the research state of the technology; it is not medical advice. Scope: In: blood-based tests for signals of several cancers at once (cell-free DNA methylation, fragmentomics, protein markers) used to screen people without symptoms. Out: tests for one cancer type, tests for people with symptoms or a known cancer, and treatment. Sequencing at scale is covered by biotech/low-cost-dna-sequencing. Metrics: - sensitivity: current 16.8%, target 50%, distance to target 33.2 points. Conditions: Stage I cancers of all types detected, blood-based methylation test, case-control validation set. Target rationale: Atlas working target, not set by an agency or a regulator: the point at which a test finds more stage I cancers than it misses. No source read for this entry sets a numeric stage I threshold; the systematic review says population screening needs high specificity and reasonable sensitivity for early-stage disease. Evidence: klein2021clinical, wade2024multi. - specificity: current 99.5%. Conditions: Cancer signal detection, case-control validation set. Evidence: klein2021clinical. Gaps: - Low sensitivity for stage I cancers [critical, scientific-unknown, layer principle, active]: In the case-control validation study sensitivity rose with stage, from 16.8% at stage I to 90.1% at stage IV, so the test finds mostly cancers that are already advanced. Stage I is where early detection would matter most and where little tumour DNA reaches the blood. Blocked by: biotech/low-cost-dna-sequencing. Evidence: klein2021clinical. - Benefit to patients not shown in a randomized trial [high, scientific-unknown, layer deployment, open]: The NHS-Galleri randomized trial of 142,250 participants reported that its primary endpoint, a reduction in stage III/IV diagnoses in the test arm, was not met. Detecting cancers is not the same as improving outcomes; longer follow-up and other trials are needed. Evidence: neal2026performance. Requires: - biotech/low-cost-dna-sequencing: Methylation and fragment-based tests read cell-free DNA by sequencing, so test price follows sequencing price. ## High-selectivity membranes (materials/high-selectivity-membranes) Status: proposed. Readiness: not assessed. Membranes that separate salt from water or one gas from another with both high permeability and high selectivity, so that separation needs energy close to the thermodynamic minimum. Scope: In: polymer, ceramic and two-dimensional-material membranes for desalination and gas separation. Out: thermal separation processes, and sorbents (materials/low-cost-co2-sorbents). Required by: water/low-energy-desalination. ## Low-cost CO2 sorbents (materials/low-cost-co2-sorbents) Status: proposed. Readiness: not assessed. Solid or liquid materials that capture CO2 from ambient air with high capacity, fast kinetics and low regeneration energy, for years of cycles, at a material cost low enough for gigatonne removal. Scope: In: amines on supports, metal-organic frameworks, hydroxide and carbonate loops for direct air capture. Out: sorbents for flue gas at 4 to 15% CO2 and membranes (materials/high-selectivity-membranes). Required by: climate/direct-air-capture. ## Low-cost high-temperature superconducting conductor (materials/low-cost-hts-conductor) Status: scoping. Readiness: not assessed. High-temperature superconducting tape (REBCO) that carries kiloamperes at fusion-magnet conditions at a price per kA-m that lets magnets of hundreds of kilometres of tape be built. Scope: In: REBCO coated conductors and their manufacturing cost. Out: low-temperature superconductors such as Nb3Sn and NbTi, and the cabling and magnet engineering built on the tape. Metrics: - conductor-cost: current 100 USD kA⁻¹ m⁻¹, target 20 USD kA⁻¹ m⁻¹, distance to target 0.7 orders of magnitude. Conditions: REBCO tape at 20 K and 20 T, field parallel to c; derived from price per metre and critical current. Target rationale: The cost target for power applications set by a 2026 cost model and roadmap for PLD REBCO tape (fumarulo2026evaluating), which also gives the roadmap of growth rate, REBCO thickness and critical current to get there. Evidence: zhao2025commercial, fumarulo2026evaluating. Required by: energy/fusion-power. ## High-bandwidth brain-computer interface (neurotech/high-bandwidth-bci) Status: scoping. Readiness: not assessed. An implanted interface that lets a person with paralysis communicate at close to the speed of natural conversation, reliably, from a device that works for years. This entry maps research; it is not medical advice. Scope: In: intracortical and other implanted interfaces that decode attempted speech or movement into text, sound or control. Out: non-invasive interfaces, and neural stimulation therapies. The recording electrodes are covered by neurotech/long-lived-neural-electrodes. Metrics: - communication-rate: current 62 words min⁻¹, target 160 words min⁻¹, distance to target 0.4 orders of magnitude. Conditions: Speech-to-text decoding of attempted speech, intracortical arrays, one participant with ALS, large vocabulary. Target rationale: The speed of natural conversation as stated in the source paper, the rate at which a spoken exchange flows without the user waiting. Evidence: willett2023high, willett2023high. Gaps: - Decoding rate is well below natural conversation [high, engineering, layer system, active]: The 2023 demonstration decoded 62 words per minute, which the authors describe as beginning to approach the speed of natural conversation, 160 words per minute. The result comes from one participant, so how it generalizes across people is not established. Blocked by: neurotech/long-lived-neural-electrodes. Evidence: willett2023high. Requires: - neurotech/long-lived-neural-electrodes: Decoding depends on recordings from implanted electrodes that must keep working for years. - ai/energy-efficient-inference: Decoding runs a neural network on the neural signal; running it in a wearable or implanted device needs low energy per inference. ## Long-lived neural electrodes (neurotech/long-lived-neural-electrodes) Status: proposed. Readiness: not assessed. Implanted electrodes that record the activity of many neurons with stable signal quality for many years, without damaging tissue or being rejected by the body. Scope: In: intracortical electrode arrays and thin-film or flexible probes, and the signal stability and tissue response over time. Out: the decoding software and the interface as a whole (neurotech/high-bandwidth-bci). Required by: neurotech/high-bandwidth-bci. ## Fault-tolerant quantum computer (quantum/fault-tolerant-quantum-computer) Status: mapped. Readiness: TRL 4 (4 of 9). Readiness note: A distance-7 surface-code memory below threshold has been shown in the laboratory; no logical algorithm of practical value has run, and the wiring, refrigeration and decoding for a million-qubit machine are not built. A quantum computer that runs algorithms of practical value, with logical error rates low enough that the answer can be trusted. It does this by encoding each logical qubit in many noisy physical qubits and correcting errors faster than they accumulate. Scope: In scope: the full machine that error-corrects a computation, built here from physical qubits, a real-time decoder, cryogenic control and refrigeration, and links between modules. Out of scope: noisy intermediate-scale machines without error correction, quantum sensors and quantum networks for their own sake. The decoder is covered by quantum/real-time-qec-decoder and the module links by quantum/quantum-interconnect. Metrics: - logical-error-per-cycle: current 0.00143, target 10⁻¹², distance to target 9.2 orders of magnitude. Conditions: Surface-code memory, distance 7, 101 qubits, superconducting processor. Target rationale: Gidney's RSA-2048 estimate runs for less than a week at a 1 microsecond cycle, about 6e11 cycles (6.05e5 s divided by 1e-6 s; durations from gidney2025how). A single logical qubit must then fail with probability well under 1/(6e11), about 1.7e-12, per cycle; with many logical qubits the requirement is lower still, so 1e-12 is a floor of the order of magnitude, not a precise budget. Evidence: google2024quantum, gidney2025how. - two-qubit-gate-infidelity: current 6 × 10⁻⁴, target 0.001, distance to target met. Conditions: Best reported single pair: 60 ns gate on two fluxonium qubits, randomized benchmarking. Target rationale: Gidney's RSA-2048 estimate assumes a uniform gate error of 0.1% across a square grid of qubits. The best pair already beats it, so the gap is uniformity across a million qubits, not the best pair. Evidence: lin2025days, gidney2025how. - physical-qubits: current 101 qubit, target 10⁶ qubit, distance to target 4.0 orders of magnitude. Conditions: Qubits used by the distance-7 surface-code memory on the Willow processor. Target rationale: Upper bound of the RSA-2048 resource estimate: less than a million noisy qubits (gidney2025how), down from 20 million in the 2019 estimate. Evidence: google2024quantum, gidney2025how. - decoder-latency: current 6.3 × 10⁻⁵ s, target 10⁻⁵ s, distance to target 0.8 orders of magnitude. Conditions: Average real-time decoder latency, distance-5 surface code, cycle time 1.1 microseconds. Target rationale: The control-system reaction time of 10 microseconds assumed by the RSA-2048 resource estimate (gidney2025how). Evidence: google2024quantum, gidney2025how. Gaps: - Physical error rates uniform across the array [high, engineering, layer device, active]: Single pairs of superconducting qubits reach two-qubit gate fidelity of 99.94% (error 6e-4) with stability over 24 days, but a million-qubit machine needs every pair near that level at once, with leakage and crosstalk held down. The Willow memory ran at 0.143% logical error per cycle at distance 7, short of the target by about nine orders of magnitude. Closing it means both lower physical error and larger code distance on many more qubits. Approach: Fluxonium qubits with direct two-qubit gates. Evidence: google2024quantum, lin2025days. - Qubit count and control wiring [critical, engineering, layer system, active]: The largest error-corrected memory in the evidence uses 101 qubits; the target is up to a million. A coaxial line per qubit from room temperature does not scale, so control and readout must move into the cryostat. Cryo-CMOS multiplexing has worked below 15 mK without degrading relaxation times, and superconducting digital demultiplexing has run a multi-qubit system, both at laboratory scale. Blocked by: enablers/cryogenic-control-electronics, enablers/dilution-refrigeration. Approach: Cryo-CMOS multiplexer below 15 mK. Approach: Superconducting digital control electronics at millikelvin. Evidence: acharya2023multiplexed, jordan2026quantum, krinner2019engineering, pauka2019cryogenic. - Correlated errors from cosmic rays and radioactivity [high, scientific-unknown, layer device, promising]: Muons and gamma rays create quasiparticle bursts that cause correlated errors across a chip, which error correction cannot absorb. A measurement on a 63-qubit processor separated the contributions of muons and gamma rays. Back-side phonon downconversion cut correlated poisoning by two orders of magnitude on three-qubit chips; it has not been shown on a million-qubit array. Approach: Phonon downconversion with back-side normal-metal reservoirs. Evidence: li2025cosmic, iaia2022phonon. - Decoding at scale and in real time [high, engineering, layer system, active]: The Willow decoder averaged 63 microseconds at distance 5, six times the 10 microsecond reaction time assumed in the RSA-2048 estimate, and at a cycle time of 1.1 microseconds, far from 1 microsecond at thousands of logical qubits. FPGA decoders report tens to hundreds of nanoseconds per measurement round in modeled-noise studies, which is not the same quantity as end-to-end latency in a running experiment. Blocked by: quantum/real-time-qec-decoder. Approach: Distributed Union-Find decoder on FPGA. Evidence: google2024quantum, liyanage2024fpga, gidney2025how. - Refrigeration for the heat load [high, engineering, layer system, open]: Each wired qubit adds passive heat load from cables and active load from signal dissipation. The strongest dilution refrigerator in the evidence delivers 2 mW at about 100 mK with a base temperature of 6.6 mK; no source in the atlas states the load of a million-qubit machine. Resource models of modular machines predict power and thermal load, and point to splitting the machine across cryostats. Blocked by: enablers/dilution-refrigeration. Approach: Modular machine split across several cryostats. Evidence: guan2025development, krinner2019engineering, saadatmand2026superconducting. - Links between modules [high, engineering, layer system, open]: A multinode machine needs entanglement between cryostats. Internode gates may be two to three orders of magnitude noisier and slower than local operations, and a systems analysis finds link performance must improve by 10 to 100 times. Blocked by: quantum/quantum-interconnect. Approach: Optical interconnects with entanglement distillation. Evidence: ang2024arquin. Impact: - Risk: A quantum computer with less than a million noisy qubits could factor a 2048-bit RSA integer in less than a week, under the estimate's assumptions of a uniform gate error of 0.1%, a surface-code cycle of 1 microsecond and a control reaction time of 10 microseconds. Who: Anyone whose data is protected by RSA-2048. Class: reported. Unlocked by the target for physical-qubits. UN SDG: 9. Evidence: gidney2025how. Requires: - quantum/real-time-qec-decoder: Every error-correction cycle needs its syndromes decoded before the next logical operation that depends on them. Need: Reaction time of 10 microseconds, the figure assumed in the RSA-2048 resource estimate (gidney2025how). - enablers/dilution-refrigeration: Superconducting qubits operate at about 10 mK, and every control line and amplifier adds heat to that stage. Need: Enough cooling power, in one cryostat or several linked, for the heat load of up to a million physical qubits and their wiring. - enablers/cryogenic-control-electronics: One coaxial line per qubit does not scale to a million qubits; control and readout electronics must sit in the cold. - quantum/quantum-interconnect: A machine of up to a million qubits is unlikely to fit in one cryostat, so modules must be linked by quantum channels. The Alan Machine: [The quantum Alan](https://the-alan-machine.github.io/the-alan-machine/chapters/quantum-alan/index.html) The Alan Machine: [Building Alan: quantum hardware](https://the-alan-machine.github.io/the-alan-machine/building-alan/quantum-hardware/index.html) ## Quantum interconnect (quantum/quantum-interconnect) Status: proposed. Readiness: not assessed. Channels that carry quantum states or entanglement between quantum processors in separate cryostats or modules, so that a machine can grow past what one cryostat holds. Scope: In scope: links between modules of a superconducting machine, such as microwave-to-optical transduction and optical entanglement distribution, and their fidelity and rate. Out of scope: long-distance quantum networks and communication for their own sake, and on-chip couplers inside a module. Requires: - enablers/photonic-integration: Optical links between modules need low-loss waveguides, modulators and detectors integrated on chip. Required by: quantum/fault-tolerant-quantum-computer. ## Real-time quantum error-correction decoder (quantum/real-time-qec-decoder) Status: scoping. Readiness: not assessed. Classical hardware and algorithms that turn the stream of syndrome measurements of an error-correcting code into corrections, fast enough that the quantum computer never waits for them. Scope: In scope: the decoding algorithm, its implementation on FPGA or ASIC, and its latency and throughput during a live experiment. Out of scope: the choice of code, the qubits themselves (quantum/fault-tolerant-quantum-computer) and the cryogenic electronics it may sit on (enablers/cryogenic-control-electronics). Metrics: - decoder-latency: current 6.3 × 10⁻⁵ s, target 10⁻⁵ s, distance to target 0.8 orders of magnitude. Conditions: Average real-time decoder latency in a running experiment: distance-5 surface code, cycle time 1.1 microseconds. Target rationale: The control-system reaction time of 10 microseconds assumed by the RSA-2048 resource estimate (gidney2025how). Evidence: google2024quantum, gidney2025how. Requires: - enablers/cryogenic-control-electronics: The decoder reads syndromes from the control system, and placing it near the qubits shortens the path and the latency. Required by: quantum/fault-tolerant-quantum-computer. ## Closed-loop life support (space/closed-loop-life-support) Status: proposed. Readiness: not assessed. A life-support system for people in space or on another world that recycles air, water and food waste so that resupply from Earth is a small fraction of the crew's consumption. Scope: In: recovery of water and oxygen, waste processing, and bioregenerative food production for a crew. Out: launch to orbit (space/low-cost-access-to-orbit), radiation shielding and the power plant. The headline metric, a closure fraction of mass flows, is not yet in taxonomy/metrics.toml, so the technology stays proposed. ## Low-cost access to orbit (space/low-cost-access-to-orbit) Status: scoping. Readiness: not assessed. Placing payload in low Earth orbit at a cost per kilogram low enough that mass is no longer the main limit on what is built in space. Scope: In: the cost per kilogram of launch to low Earth orbit, by reusable or otherwise cheaper launch systems. Out: what is built or kept alive in orbit (space/closed-loop-life-support) and in-space transport beyond low Earth orbit. Metrics: - launch-cost: current 3 868 USD kg⁻¹, target 300 USD kg⁻¹, distance to target 1.1 orders of magnitude. Conditions: Average cost of sending one kilogram to orbit, all launches in the year. Target rationale: The central-estimate projection of the same experience-curve study for 2040; it is a forecast from a Wright's-law fit, not a goal set by an agency, and is used here as the next milestone. Evidence: terzi2026from, terzi2026from. Gaps: - Reuse of the whole launch system [high, cost, layer deployment, active]: Reusable first stages reduced launch cost, and providers on several continents plan reusable systems. The cost per kilogram falls with cumulative payload launched, following a learning curve, so further reductions depend on reuse at higher rate and with more of the vehicle recovered. Evidence: reddy2018spacex, terzi2026from. - Market structure and orbital debris may slow the decline [medium, regulation, layer deployment, open]: The study that fits the learning curve warns that geopolitical shifts, possible monopolistic behavior in commercial launch markets and the growing problem of orbital debris may temper the cost reductions. Evidence: terzi2026from. ## Low-energy desalination (water/low-energy-desalination) Status: scoping. Readiness: not assessed. Turning seawater into fresh water using electricity close to the thermodynamic minimum for the separation, so that desalination is no longer limited by its energy bill. Scope: In: seawater desalination by membranes (reverse osmosis, electrodialysis) and the energy recovery and staging around them. Out: the membranes themselves, which are materials/high-selectivity-membranes; brackish and wastewater reuse, which start from much lower salinity and need less energy; and the power source, which belongs to the energy domain. Metrics: - specific-energy-consumption: current 3 kWh m⁻³, target 1 kWh m⁻³, distance to target 0.5 orders of magnitude. Conditions: Seawater feed, electricity per cubic metre of fresh water. Target rationale: A round value chosen by the atlas, not set by an agency: about the level a simulation study reached for reverse osmosis on osmotically diluted seawater (0.96 kWh/m3), and a third of the current value, so that energy stops dominating the cost of desalinated water. Evidence: doornbusch2021multistage, yalamanchili2024can. Gaps: - Practical plants already work near the thermodynamic limit [high, fundamental-limit, layer principle, open]: Reversible reverse osmosis and electrodialysis consume the Gibbs free energy of separation, and the practical energy of both approaches that minimum only as the number of stages grows. Reviews state that most desalination technologies already work near their limit, so the remaining reduction is bounded and each step costs capital. The numeric value of the minimum is in the body of the cited papers and is not recorded here yet. Evidence: wang2020derivation, elimelech2011future, nassrullah2020energy. - Membrane permeability and selectivity trade off [high, engineering, layer device, active]: Reverse osmosis membranes trade water permeability against salt rejection. Analysis of biomimetic membranes indicates what a more permeable, defect-free membrane could offer for seawater desalination, but such membranes are not yet a commercial product. Blocked by: materials/high-selectivity-membranes. Evidence: werber2018permselectivity, elimelech2011future. Requires: - materials/high-selectivity-membranes: Membrane permeability and selectivity set how close staged desalination can get to the thermodynamic minimum. # Evidence cards ## abushawareb2024achievement Achievement of Target Gain Larger than Unity in an Inertial Fusion Experiment. Abu-Shawareb, H.. Physical Review Letters, 2024. Source: https://doi.org/10.1103/PhysRevLett.132.065102 Class: established. Status: machine-checked. Cited by: energy/fusion-power. - scientific-gain = 1.5 1. Conditions: Indirect-drive implosion at the National Ignition Facility, 2.05 MJ of 351 nm laser light, 3.1 MJ fusion yield, 2022-12-05. Quote: "an indirect drive fusion implosion on the National Ignition Facility (NIF) achieved a target gain G_{target} of 1.5" ## acharya2023multiplexed Multiplexed superconducting qubit control at millikelvin temperatures with a low-power cryo-CMOS multiplexer. Acharya, R., Brebels, S., Grill, Alexander, Verjauw, Jeroen, Ivanov, Ts., Lozano, Daniel Pérez, Wan, Danny, Van Damme, Jacques. Nature Electronics, 2023. Source: https://doi.org/10.1038/s41928-023-01033-8 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Radio-frequency cryo-CMOS multiplexer operating below 15 mK, interfaced with a superconducting qubit. Quote: "operates below 15 mK with a minimal cross-coupling" ## allen2017expression Expression of 16 Nitrogenase Proteins within the Plant Mitochondrial Matrix. Allen, Robert S., Tilbrook, Kimberley, Warden, Andrew C., Campbell, Peter C., Rolland, Vivien, Singh, Surinder Pal, Wood, Craig C.. Frontiers in Plant Science, 2017. Source: https://doi.org/10.3389/fpls.2017.00287 Class: established. Status: machine-checked. Cited by: food/nitrogen-fixing-cereals. - Conditions: Qualitative statement from the abstract; expression in Nicotiana benthamiana, not a crop. Quote: "the NifD catalytic component was the least abundant" ## almogy2022cost Cost-efficient whole genome-sequencing using novel mostly natural sequencing-by-synthesis chemistry and open fluidics platform. Almogy, Gilad, Pratt, Mark, Oberstrass, Florian C., Lee, Linda, Mazur, Dan, Beckett, Nate, Barad, Omer, Soifer, Ilya. bioRxiv, 2022. Source: https://doi.org/10.1101/2022.05.29.493900 Class: reported. Status: machine-checked. Cited by: biotech/low-cost-dna-sequencing. - sequencing-cost = 90 USD. Conditions: Derived, not stated in the abstract: 1 USD per Gb times 30-fold coverage of a 3 Gb human genome (3,000 Mb and 30-fold coverage are the assumptions NHGRI uses for Illumina-type platforms, see wetterstrand2023dna). Cost basis of the 1 USD/Gb figure is not defined in the abstract. Quote: "at a low cost of $1/Gb" ## ang2024arquin ARQUIN: Architectures for Multinode Superconducting Quantum Computers. Ang, James, Carini, Gabriella A., Chen, Yanzhu, Chuang, Isaac L., Demarco, Michael, Economou, Sophia E., Eickbusch, Alec, Faraon, Andrei. ACM Transactions on Quantum Computing, 2024. Source: https://doi.org/10.1145/3674151 Class: extrapolation. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Systems analysis of multinode superconducting computers with optical interconnects. Quote: "internode gates in these systems may be two to three orders of magnitude noisier and slower than local operations" - Conditions: Same analysis; the link performance the algorithms require. Quote: "We find that a factor of 10–100× better link performance is required" ## barcohen2019embedded Embedded Cooling for Wide Bandgap Power Amplifiers: A Review. Bar‐Cohen, Avram, Maurer, J. J., Altman, David H.. Journal of Electronic Packaging, 2019. Source: https://doi.org/10.1115/1.4043404 Class: established. Status: machine-checked. Cited by: enablers/heat-removal. - heat-flux = 400000000.0 W m^-2. Conditions: Record heat flux above 40 kW/cm2 (4e8 W/m2) in DARPA's near-junction thermal transport program, GaN-on-diamond, embedded cooling, small heated area. Quote: "heat fluxes, above 40 kW/cm2, achieved in Defense Advanced Research Projects Agency (DARPA)'s near-junction thermal transport (NJTT) program" ## bennett2023improving Improving de novo protein binder design with deep learning. Bennett, Nathaniel R., Coventry, Brian, Goreshnik, Inna, Huang, Buwei, Allen, Aza, Vafeados, Dionne K., Peng, Ying Po, Dauparas, Justas. Nature Communications, 2023. Source: https://doi.org/10.1038/s41467-023-38328-5 Class: established. Status: machine-checked. Cited by: biotech/de-novo-protein-design. - Conditions: Deep-learning filtering of designed binders (AlphaFold2 or RoseTTAFold) against energy-based design. Quote: "increases design success rates nearly 10-fold" ## bose2024anneal Anneal-free ultra-low loss silicon nitride integrated photonics. Bose, Debapam, Harrington, Mark, Isichenko, Andrei, Liu, Kaikai, Wang, Jiawei, Chauhan, Nitesh, Newman, Zachary L., Blumenthal, Daniel J.. Light Science & Applications, 2024. Source: https://doi.org/10.1038/s41377-024-01503-4 Class: established. Status: machine-checked. Cited by: enablers/photonic-integration. - waveguide-loss = 1.77 dB m^-1. Conditions: 80 nm nitride core waveguides, anneal-free process at a maximum of 250 C. Quote: "enabling 1.77 dB m -1 loss and 14.9 million Q for 80 nm nitride core waveguides" ## doornbusch2021multistage Multistage electrodialysis for desalination of natural seawater. Doornbusch, Gijs, van der Wal, Marrit, Tedesco, Michele, Post, Jan W., Nijmeijer, Kitty, Borneman, Zandrie. Desalination, 2021. Source: https://doi.org/10.1016/j.desal.2021.114973 Class: established. Status: machine-checked. Cited by: water/low-energy-desalination. - specific-energy-consumption = 3 kWh m^-3. Conditions: Measured, upscaled multistage electrodialysis on natural seawater, 27 g/l in and 1.9 g/l out, stable over 18 days. Electrodialysis, not reverse osmosis. Quote: "The system performance was stable over 18 days, with an average energy consumption of 3 kWh/m3" ## elimelech2011future The Future of Seawater Desalination: Energy, Technology, and the Environment. Elimelech, Menachem, Phillip, William A.. Science, 2011. Source: https://doi.org/10.1126/science.1200488 Class: established. Status: machine-checked. Cited by: water/low-energy-desalination. - Conditions: Qualitative statement from the abstract of a review; no metric value. Quote: "seawater desalination is still more energy intensive compared to conventional technologies for the treatment of fresh water" ## elsworth2025measuring Measuring the environmental impact of delivering AI at Google Scale. Elsworth, Cooper, Huang, Keguo, Patterson, David, Schneider, Ian, Sedivy, Robert, Goodman, Savannah, Townsend, Ben, Ranganathan, Parthasarathy. arXiv, 2025. Source: https://arxiv.org/abs/2508.15734 Class: reported. Status: machine-checked. Cited by: ai/energy-efficient-inference. - Conditions: Per prompt, not per token: median Gemini Apps text prompt, full serving stack (0.24 Wh, about 864 J). No metric of the atlas is per prompt, so this finding carries no metric. Quote: "the median Gemini Apps text prompt consumes 0.24 Wh of energy" ## fumarulo2026evaluating Evaluating the scalability of REBCO coated-conductor manufacturing by pulsed laser deposition. Fumarulo, Savino, Massa, Leonardo, Sesti, Valentina, Kirkels, Arjan, Senatore, Carmine, Martina, Mario L V. Superconductor Science and Technology, 2026. Source: https://doi.org/10.1088/1361-6668/aea455 Class: established. Status: machine-checked. Cited by: materials/low-cost-hts-conductor. - conductor-cost = 20 USD kA^-1 m^-1. Conditions: Cost target for power applications stated by the authors; not a measured price. Quote: "we present a technology roadmap to reach the cost target of $20/(kA·m) required for power applications" ## gidney2025how How to factor 2048 bit RSA integers with less than a million noisy qubits. Gidney, Craig. arXiv, 2025. Source: https://arxiv.org/abs/2505.15917 Class: reported. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer, quantum/real-time-qec-decoder. - physical-qubits = 1000000.0 qubit. Conditions: Upper bound on qubits for factoring a 2048 bit RSA integer in less than a week; the abstract says less than a million, so 1e6 is the bound, not an estimate of the exact count. Quote: "I estimate that a 2048 bit RSA integer could be factored in less than a week by a quantum computer with less than a million noisy qubits." ## goncharov2025laser Laser requirements for inertial fusion energy target designs. Goncharov, Valeri N.. Optical Technologies for Inertial Fusion Energy, 2025. Source: https://doi.org/10.1117/12.3041301 Class: reported. Status: machine-checked. Cited by: energy/fusion-power. - scientific-gain = 100 1. Conditions: Requirement for an inertial fusion energy plant, not a measurement: neutron yield over incident laser energy. Quote: "implosion physics can meet the high-gain requirements for IFE (when a ratio of neutron yield to incident laser energy ~100)" ## google2024quantum Quantum error correction below the surface code threshold. Google Quantum AI and Collaborators. Nature, 2024. Source: https://doi.org/10.1038/s41586-024-08449-y Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer, quantum/real-time-qec-decoder. - logical-error-per-cycle = 0.00143 1. Conditions: Distance-7 surface-code memory on 101 qubits of the Willow superconducting processor. Quote: "culminating in a 101-qubit distance-7 code with 0.143% ± 0.003% error per cycle of error correction" - decoder-latency = 6.3e-05 s. Conditions: Average real-time decoder latency, distance-5 surface code, up to a million cycles of 1.1 microseconds each. Quote: "achieving an average decoder latency of 63 μs at distance-5 up to a million cycles, with a cycle time of 1.1 μs" - physical-qubits = 101 qubit. Conditions: Qubits used by the distance-7 surface-code memory. Quote: "culminating in a 101-qubit distance-7 code" ## guan2025development Development of a cryogen-free dilution refrigerator with a cooling power of 2000 μ W at around 100 mK. Guan, Xiang, Wang, De Ming, Xie, Yong Jie, Fan, Jie, Ji, Zhong Qing. Review of Scientific Instruments, 2025. Source: https://doi.org/10.1063/5.0294894 Class: established. Status: machine-checked. Cited by: enablers/dilution-refrigeration, quantum/fault-tolerant-quantum-computer. - cooling-power = 0.002 W. Conditions: Cryogen-free dilution refrigerator, four parallel dilution units, two pulse-tube precoolers; at 102.48 mK; 2000 microwatts converted to 2e-3 W. Quote: "a cooling power of 2000 μW at 102.48 mK" ## guo2022biological Biological nitrogen fixation in cereal crops: Progress, strategies, and perspectives. Guo, Kaiyan, Yang, Jun, Yu, Nan, Luo, Li, Wang, Ertao. Plant Communications, 2022. Source: https://doi.org/10.1016/j.xplc.2022.100499 Class: established. Status: machine-checked. Cited by: food/nitrogen-fixing-cereals. - Conditions: Qualitative statement from the abstract of a review. Quote: "Engineering cereal crops that can fix nitrogen like legumes or associate with nitrogen-fixing microbiomes could help to avoid the problems caused by the overuse of synthetic nitrogen fertilizer." ## ho2023limits Limits to the Energy Efficiency of CMOS Microprocessors. Ho, Anson, Erdil, Ege, Besiroglu, Tamay. arXiv, 2023. Source: https://arxiv.org/abs/2312.08595 Class: extrapolation. Status: machine-checked. Cited by: ai/energy-efficient-inference. - energy-efficiency = 4700000000000000.0 FLOP J^-1. Conditions: Geometric-mean estimate of the maximum floating point operations per joule for CMOS microprocessors, counted in FP4 operations, combining transistor switching, interconnect capacitance and leakage. An estimate of a ceiling, not a current value. Quote: "Combining these yields a geometric mean estimate of 4.7e15 FP4/J for the maximum CMOS energy efficiency, roughly two hundred-fold more efficient than current microprocessors." ## huisman2019roadmap A Roadmap toward Engineered Nitrogen-Fixing Nodule Symbiosis. Huisman, Rik, Geurts, René. Plant Communications, 2019. Source: https://doi.org/10.1016/j.xplc.2019.100019 Class: established. Status: machine-checked. Cited by: food/nitrogen-fixing-cereals. - Conditions: Qualitative statement from the abstract of a review. Quote: "the long-standing objective to engineer the nitrogen-fixing nodulation trait on non-leguminous crop plants has not been achieved yet" ## iaia2022phonon Phonon downconversion to suppress correlated errors in superconducting qubits. Iaia, Vito, Ku, Jaseung, Ballard, A., Larson, C. P., Yelton, E., Liu, C. H., Patel, S., McDermott, Robert. Nature Communications, 2022. Source: https://doi.org/10.1038/s41467-022-33997-0 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Back-side normal-metal phonon downconversion on qubit chips, three qubits monitored per chip. Quote: "observe a two-order of magnitude reduction in correlated poisoning due to background radiation" ## jiang2025rfdiffusion RFdiffusion Exhibits Low Success Rate in De Novo Design of Functional Protein Binders for Biochemical Detection. Jiang, Bruce, Li, Xiaoxiao, Guo, Amber, Wei, Moris, Wu, Junhua. bioRxiv, 2025. Source: https://doi.org/10.1101/2025.02.07.636769 Class: reported. Status: machine-checked. Cited by: biotech/de-novo-protein-design. - Conditions: Six targets, five RFdiffusion designs each, tested by one group. Quote: "Binders for the other targets failed due to low expression, nonspecific binding, or undetectable affinity" ## jordan2026quantum A quantum computer controlled by superconducting digital electronics at millikelvin temperature. Jordan, Caleb, Bernhardt, Jacob, Rahamim, Joseph, Kirichenko, Alex, Bharadwaj, Karthik Srikanth, Fry-Bouriaux, Louis, Somoroff, Aaron, Porsch, Katie. Nature Electronics, 2026. Source: https://doi.org/10.1038/s41928-026-01576-6 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Multi-qubit system with superconducting digital demultiplexing of control lines. Quote: "The system utilizes digital demultiplexing, breaking the linear scaling of control lines to number of qubits." ## keith2018process A Process for Capturing CO2 from the Atmosphere. Keith, David W.. Joule, 2018. Source: https://doi.org/10.1016/j.joule.2018.05.006 Class: extrapolation. Status: machine-checked. Cited by: climate/direct-air-capture. - capture-cost = 94 USD t^-1. Conditions: Lower end of the levelized cost range of a designed aqueous-KOH plant capturing about 1 Mt-CO2 per year, with CO2 delivered at 15 MPa; capture only, storage not included. Upper end 232 USD/t. Design estimate, not a current value. Quote: "the levelized cost per ton CO 2 captured from the atmosphere ranges from 94 to 232 $/t-CO 2" ## klein2021clinical Clinical validation of a targeted methylation-based multi-cancer early detection test using an independent validation set. Klein, Eric A., Richards, Donald, Cohn, Allen Lee, Tummala, Meghna, Lapham, Rosanna L., Cosgrove, David Owen, Chung, Gina G., Clement, Jessica. Annals of Oncology, 2021. Source: https://doi.org/10.1016/j.annonc.2021.05.806 Class: established. Status: machine-checked. Cited by: health/multi-cancer-early-detection. - sensitivity = 16.8 %. Conditions: Stage I cancers, any of more than 50 cancer types, blood-based methylation MCED test, independent validation set of 4077 participants (case-control). Quote: "stage I: 16.8% (14.5% to 19.5%)" - specificity = 99.5 %. Conditions: Cancer signal detection in the independent validation set, 1254 non-cancer participants. Quote: "Specificity for cancer signal detection was 99.5%" ## krinner2019engineering Engineering cryogenic setups for 100-qubit scale superconducting circuit systems. Krinner, Sebastian, Storz, Simon, Kurpiers, Philipp, Magnard, Paul, Heinsoo, Johannes, Keller, Raphael, Lütolf, J., Eichler, Christopher. EPJ Quantum Technology, 2019. Source: https://doi.org/10.1140/epjqt/s40507-019-0072-0 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Measured passive heat load of stainless steel and NbTi coaxial cables and active load of signal dissipation. Quote: "The passive heat load of stainless steel and NbTi coaxial cables and the active load due to signal dissipation are measured" ## li2025cosmic Cosmic-ray-induced correlated errors in superconducting qubit array. Li, Xuegang, Wang, Junhua, Jiang, Yao-Yao, Xue, Guangming, Cai, Xiaoxia, Zhou, Jun, Gong, Ming, Liu, Zhaofeng. Nature Communications, 2025. Source: https://doi.org/10.1038/s41467-025-59778-z Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Direct observation of muon-induced quasiparticle bursts and correlated errors on a 63-qubit processor, with muon detectors in the dilution refrigerator. Quote: "We directly observe QP bursts leading to correlated errors that are induced solely by muons and separate the contributions of muons and γ-rays." ## lin2025days 24 Days-Stable CNOT Gate on Fluxonium Qubits with Over 99.9% Fidelity. Lin, Wei-Ju, Cho, Hyunheung, Chen, Yinqi, Vavilov, Maxim, Wang, Chen, Manucharyan, Vladimir. PRX Quantum, 2025. Source: https://doi.org/10.1103/prxquantum.6.010349 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - two-qubit-gate-infidelity = 0.0006 1. Conditions: 60 ns two-qubit gate on one pair of fluxonium qubits, randomized benchmarking; average gate fidelity 99.94%, so infidelity 1 - 0.9994 = 6e-4. Quote: "estimated using randomized benchmarking, was as high as 99.94%" ## liyanage2024fpga FPGA-Based Distributed Union-Find Decoder for Surface Codes. Liyanage, Namitha, Wu, Yue, Tagare, Siona, Zhong, Lin. IEEE Transactions on Quantum Engineering, 2024. Source: https://doi.org/10.1109/tqe.2024.3467271 Class: established. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Xilinx VCU129 FPGA decoding distance 51 under phenomenological noise; average latency per measurement round. Quote: "decoding d=51 on a Xilinx VCU129 FPGA with an average latency of 544 ns per measurement round" ## mcgeoch2025development Development path to an ArF laser-fusion pilot power plant. McGeoch, Malcolm W., Obenschain, Stephen P.. Optical Technologies for Inertial Fusion Energy, 2025. Source: https://doi.org/10.1117/12.3047795 Class: reported. Status: machine-checked. Cited by: energy/fusion-power. - scientific-gain = 100 1. Conditions: Requirement for a laser-fusion power plant, from simulations at the Naval Research Laboratory; not a measurement. Quote: "the gains (>100) needed for a laser-fusion power plant can be achieved with less than 1MJ of laser energy" ## nassrullah2020energy Energy for desalination: A state-of-the-art review. Nassrullah, Haya, Anis, Shaheen Fatima, Hashaikeh, Raed, Hilal, Nidal. Desalination, 2020. Source: https://doi.org/10.1016/j.desal.2020.114569 Class: established. Status: machine-checked. Cited by: water/low-energy-desalination. - Conditions: Qualitative statement from the abstract of a review. Quote: "Most technologies are already working near their thermodynamic limit, while posing challenges in further SEC reductions." ## neal2026performance Performance of a multi-cancer early detection test in the randomized controlled NHS-Galleri trial. Neal, Richard D, Dolly, Saoirse Olivia, Johnson, Peter, Jones, Helen, Kumar, Sir Harpal, Lee, Lennard Y. W., Lee, Yujin, Liang, Wei. Nature Medicine, 2026. Source: https://doi.org/10.1038/s41591-026-04652-8 Class: established. Status: machine-checked. Cited by: health/multi-cancer-early-detection. - Conditions: NHS-Galleri randomized trial, 142,250 participants aged 50-77. Quote: "the primary endpoint of a reduction in the incidence of stage III/IV cancer diagnoses in the intervention arm versus control arm was not met" ## oviedo2026energy Energy use of AI inference, efficiency pathways, and test-time scaling. Oviedo, Felipe, Kazhamiaka, Fiodar, Choukse, Esha, Kim, Allen, Luers, Amy Lynd, Nakagawa, Melanie, Bianchini, Ricardo, Lavista Ferres, Juan. Joule, 2026. Source: https://doi.org/10.1016/j.joule.2026.102430 Class: established. Status: machine-checked. Cited by: ai/energy-efficient-inference. - Conditions: Per query, not per token: bottom-up model estimate for frontier-scale models (>200B parameters) on H100 nodes (0.31 Wh, about 1128 J). No metric of the atlas is per query, so this finding carries no metric. Quote: "we estimate a median energy of 0.31 Wh/query" ## pacesa2025one One-shot design of functional protein binders with BindCraft. Pacesa, Martin, Nickel, Lennart, Schellhaas, Christian, Schmidt, Joseph H., Pyatova, Ekaterina, Kissling, Lucas, Barendse, Patrick, Choudhury, Jagrity. Nature, 2025. Source: https://doi.org/10.1038/s41586-025-09429-6 Class: established. Status: machine-checked. Cited by: biotech/de-novo-protein-design. - experimental-hit-rate = 10 %. Conditions: De novo protein binder design with an AlphaFold2-based pipeline; the range of experimental success rates across the targets tested, lower end. Quote: "experimental success rates of 10–100%" ## park2024pjbit A 0.65-pJ/bit 3.6-TB/s/mm I/O Interface with XTalk Minimizing Affine Signaling for Next-Generation HBM with High Interconnect Density. Park, Hyunjun, Shin, Ji-Won, Kim, Hanseok, Kim, Ji‐Hee, Shin, Haengbeom, Kim, Tae‐Hoon, Park, Jung-Hun, Choi, Woo‐Seok. arXiv, 2024. Source: https://arxiv.org/abs/2404.05119 Class: reported. Status: machine-checked. Cited by: ai/energy-efficient-inference, computing/low-energy-data-movement. - energy-per-bit = 6.5e-13 J. Conditions: Die-to-die I/O transceiver over silicon interposer or similar high-density interconnect, 28 nm CMOS prototype, edge density 3.6 TB/s/mm. 0.65 pJ/b converted to joules. Quote: "the prototype XMAS transceiver achieves an edge density of 3.6TB/s/mm and an energy efficiency of 0.65pJ/b" ## pauka2019cryogenic A Cryogenic Interface for Controlling Many Qubits. Pauka, S. J., Das, K., Kalra, R., Moini, A., Yang, Y., Trainer, M., Bousquet, A., Cantaloube, C.. arXiv, 2019. Source: https://arxiv.org/abs/1912.01299 Class: reported. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: CMOS control platform of 100,000 transistors near 100 mK, benchmarked on a quantum dot test device. Quote: "showing that the control of thousands of gate electrodes is feasible within the cooling power of commercially available dilution refrigerators" ## reddy2018spacex The SpaceX Effect. Reddy, Vidya Sagar. New Space, 2018. Source: https://doi.org/10.1089/space.2017.0032 Class: established. Status: machine-checked. Cited by: space/low-cost-access-to-orbit. - Conditions: Qualitative statement from the abstract of a survey. Quote: "reusability is fast becoming the norm of the launch vehicle industry" ## saadatmand2026superconducting Superconducting qubits in the millions: The potential and limitations of modularity. Saadatmand, S. N.. Physical Review Applied, 2026. Source: https://doi.org/10.1103/k3d5-v43c Class: extrapolation. Status: machine-checked. Cited by: quantum/fault-tolerant-quantum-computer. - Conditions: Resource-estimation model of a modular superconducting fault-tolerant computer with coherent links. Quote: "Our tool can predict the size, power consumption, and execution time of these algorithms based on explicit assumptions about the physical layout, thermal load, and modular connectivity of the system." ## sievert2024considering Considering technology characteristics to project future costs of direct air capture. Sievert, Katrin. Joule, 2024. Source: https://doi.org/10.1016/j.joule.2024.02.005 Class: extrapolation. Status: machine-checked. Cited by: climate/direct-air-capture. - net-removal-cost = 374 USD t^-1. Conditions: Projection, solid sorbent DACCS, cost of CO2 net removed at 1 Gt-CO2/year cumulative capacity; 90% confidence range 281 to 579 USD/t. Not a current value. Quote: "At 1 Gt-CO2/year cumulative capacity, we project DACCS costs at $341/tCO2 ($226–$544 at 90% confidence) for liquid solvent DACCS, $374/tCO2 ($281–$579) for solid sorbent DACCS" ## terzi2026from From Sputnik to Starship: Estimating the experience curve of space launch technology. Terzi, Alessio, Nicoli, Francesco. PNAS Nexus, 2026. Source: https://doi.org/10.1093/pnasnexus/pgag217 Class: established. Status: machine-checked. Cited by: space/low-cost-access-to-orbit. - launch-cost = 3868 USD kg^-1. Conditions: Average cost of sending one kilogram to orbit across all launches in 2025, from a standardized dataset of more than 4,400 launches; an average over all vehicles, not the price of the cheapest vehicle. Quote: "the average cost of sending a kilogram to orbit has dropped from 87,023 USD in 1960 to 3,868 USD in 2025" - launch-cost = 300 USD kg^-1. Conditions: Central-estimate projection for 2040 from a Wright's-law experience curve; a projection, not a measurement. Quote: "the average cost is expected to fall to 1,600 USD/kg by 2030 and 300 USD/kg by 2040" ## vandeynze2018nitrogen Nitrogen fixation in a landrace of maize is supported by a mucilage-associated diazotrophic microbiota. Van Deynze, Allen, Zamora, Pablo, Delaux, Pierre-Marc, Heitmann, Cristobal, Jayaraman, Dhileepkumar, Rajasekar, Shanmugam, Graham, Danielle, Maëda, Junko. PLoS Biology, 2018. Source: https://doi.org/10.1371/journal.pbio.2006352 Class: established. Status: machine-checked. Cited by: food/nitrogen-fixing-cereals. - nitrogen-derived-from-atmosphere = 82 %. Conditions: Upper end of the 29-82% range measured by 15N field experiments over 5 years in an indigenous Sierra Mixe maize landrace in nitrogen-depleted soil, not a modern high-yield cultivar. Quote: "atmospheric nitrogen fixation contributed 29%-82% of the nitrogen nutrition of Sierra Mixe maize" ## vellaisamy2026characterization Characterization of Request and Token Energy Costs for LLM Inference Workloads on GPU Platforms. Vellaisamy, Prabhu, Lam, Vanessa, Blanton, Shawn, Shen, John Paul. arXiv, 2026. Source: https://arxiv.org/abs/2608.28044 Class: reported. Status: machine-checked. Cited by: ai/energy-efficient-inference. - energy-per-token = 0.72 J. Conditions: Llama-3.2-1B (dense, 1B parameters) on an NVIDIA H200 GPU, batch size 16, context 4K tokens, 512 output tokens. At 10 output tokens the same setup gives 7.46 J/token. A 1B model is far smaller than frontier models. Quote: "increasing output length from 10 to 512 tokens reduces token energy from 7.46 to 0.72 J/token" ## wade2024multi Multi-cancer early detection tests for general population screening: a systematic literature review. Wade, Ros, Nevitt, Sarah J, Liu, Yiwen, Harden, Melissa, Khouja, Claire, Raine, Gary, Churchill, Rachel C., Dias, Sofia. medRxiv, 2024. Source: https://doi.org/10.1101/2024.02.14.24302576 Class: reported. Status: machine-checked. Cited by: health/multi-cancer-early-detection. - Conditions: Systematic review of MCED tests for population screening. Quote: "The use of an MCED test for population screening requires a high specificity and a reasonable sensitivity to detect early-stage disease" ## wang2020derivation Derivation of the Theoretical Minimum Energy of Separation of Desalination Processes. Wang, Li Ares, Violet, Camille, DuChanois, Ryan M., Elimelech, Menachem. Journal of Chemical Education, 2020. Source: https://doi.org/10.1021/acs.jchemed.0c01194 Class: established. Status: machine-checked. Cited by: water/low-energy-desalination. - Conditions: Qualitative statement from the abstract; the minimum is the Gibbs free energy of separation, its value is in the body of the paper. Quote: "the energy consumption of these processes approaches the minimum thermodynamic limit with increased process staging" ## wei2026microfluidic Microfluidic cooling for high-heat-flux chips: Thermal-path compression, bottleneck migration and near-junction limits. Wei, Jie, Lin, Shuyuan, Fu, Junhao, Zhao, Zhangchi, Wei, Ning. Thermo-X, 2026. Source: https://doi.org/10.70401/tx.2026.0024 Class: established. Status: machine-checked. Cited by: ai/energy-efficient-inference, enablers/heat-removal. - heat-flux = 10000000.0 W m^-2. Conditions: Upper end of the range the review gives for microfluidic cooling, 10^2 to 10^3 W/cm2 (1e6 to 1e7 W/m2); 10^3 W/cm2 converted to W/m2. Heated area, coolant and temperature criterion vary across the cited work. Quote: "Microfluidic cooling can support 102-103 W/cm2 heat fluxes and higher local loads" ## wen2021enabling Enabling Biological Nitrogen Fixation for Cereal Crops in Fertilized Fields. Wen, Amy, Havens, Keira L., Bloch, Sarah E., Shah, Neal, Higgins, Douglas A., Davis-Richardson, Austin G., Sharon, Judee, Rezaei, Farzaneh. ACS Synthetic Biology, 2021. Source: https://doi.org/10.1021/acssynbio.1c00049 Class: established. Status: machine-checked. Cited by: food/nitrogen-fixing-cereals. - Conditions: Qualitative statement from the abstract; an engineered Kosakonia strain, in nitrogen-rich environments. Quote: "increasing nitrogen fixation activity 122-fold in nitrogen-rich environments" ## werber2018permselectivity Permselectivity limits of biomimetic desalination membranes. Werber, Jay R., Elimelech, Menachem. Science Advances, 2018. Source: https://doi.org/10.1126/sciadv.aar8266 Class: established. Status: machine-checked. Cited by: water/low-energy-desalination. - Conditions: Qualitative statement from the abstract; the numeric limits are in the body of the paper. Quote: "Defect-free biomimetic membranes thus offer great potential for seawater desalination and ultrapure water production" ## wetterstrand2023dna DNA Sequencing Costs: Data. Wetterstrand, Kris A.. NHGRI Genome Sequencing Program, 2023. Source: https://www.genome.gov/about-genomics/fact-sheets/DNA-Sequencing-Costs-Data Class: established. Status: machine-checked. Cited by: biotech/low-cost-dna-sequencing. - Conditions: NHGRI cost-per-genome definition. Quote: "the assumed genome size was 3,000 Mb (i.e., the size of a human genome)" ## willett2023high A high-performance speech neuroprosthesis. Willett, Francis R., Kunz, Erin M., Fan, Chaofei, Avansino, Donald T., Wilson, Guy H, Choi, Eun Young, Kamdar, Foram B., Glasser, Matthew F.. Nature, 2023. Source: https://doi.org/10.1038/s41586-023-06377-x Class: established. Status: machine-checked. Cited by: neurotech/high-bandwidth-bci. - communication-rate = 62 words min^-1. Conditions: Speech-to-text decoding of attempted speech from intracortical microelectrode arrays, one participant, 125,000-word vocabulary at 23.8% word error rate. Quote: "Our participant’s attempted speech was decoded at 62 words per minute" ## xiang2022silicon Silicon nitride passive and active photonic integrated circuits: trends and prospects. Xiang, Chao, Jin, Warren, Bowers, John Edward. Photonics Research, 2022. Source: https://doi.org/10.1364/prj.452936 Class: established. Status: machine-checked. Cited by: enablers/photonic-integration. - Conditions: Review of hybrid and heterogeneous integration of III-V with silicon nitride. Quote: "the integration of Si and III-V materials has enabled new large-scale, advanced silicon nitride-based photonic integrated circuits with versatile functionality" ## yalamanchili2024can Can a forward osmosis-reverse osmosis hybrid system achieve 90 % wastewater recovery and desalination energy below 1 kWh/m3? A design and simulation study. Yalamanchili, Rajashree, Rodríguez-Roda, Ignasi, Galizia, Albert, Blandin, Gaëtan. Desalination, 2024. Source: https://doi.org/10.1016/j.desal.2024.117767 Class: extrapolation. Status: machine-checked. Cited by: water/low-energy-desalination. - specific-energy-consumption = 0.96 kWh m^-3. Conditions: Software simulation (not a measurement) of reverse osmosis on seawater diluted by forward osmosis with wastewater, to 7.4 g/l. Quote: "enabling RO desalination at 0.96 kWh/m3" ## zhao2025commercial Commercial compact fusion triggered REBCO tape industry: Pulsed laser deposition technology opportunities and challenges. Zhao, Yue. Superconductivity, 2025. Source: https://doi.org/10.1016/j.supcon.2025.100188 Class: established. Status: machine-checked. Cited by: materials/low-cost-hts-conductor. - conductor-cost = 100 USD kA^-1 m^-1. Conditions: Derived: about 20 USD per metre divided by a critical current above 200 A for a 4 mm wide tape at 20 K, 20 T, field parallel to c. The critical current is a lower bound, so the true cost per kA-m is at or below this value. PLD-based REBCO tape, 2025. Quote: "PLD-REBCO tapes have demonstrated excellent in-field performance ( I c >200 A-4 mm @20K, 20T, B//c) and competitive pricing ( ∼ $20/meter)"