# MS Lattice Search in bits. Build in atoms. MS Lattice is a MatterSpace product. MatterSpace’s company motto is “Advancing Bits to Master Atoms”. MS Lattice is the first of MatterSpace’s discovery engines for science and engineering. Name what the material must do. MS Lattice finds the crystal, computes what it can do, verifies it with first-principles physics and proves it blind. MS Lattice is the engine. Your AI agent is the driver. It runs on MatterSpace’s proprietary runtime intelligence, which gets sharper with every evaluation. Overview: https://www.matterspace.ai/lattice ## Proven blind. We sealed the targets away and handed over nothing but the brief. Both came back, down to the arrangement of their atoms. Four compounds it had never seen, CaO, SrF₂, BaTiO₃ and Ni₃Al, came back too, to within a few hundredths of an ångström. - 2 of 2 sealed targets found blind - 4 of 4 unseen compounds rebuilt, within 0.016–0.037 Å - 13 of 15 right structures ranked first, from a formula alone - 73 of 73 candidates rebuilt from their record Blind rediscovery: https://www.matterspace.ai/lattice/blind-rediscovery ## One sentence in. A shortlist out. Five steps from your words to a ranked shortlist, with the proof attached. A miss stops on the spot. A candidate on its way through MS Lattice: you say what the material must do and set the limits nothing may cross, MS Lattice proposes, learns and proposes again, physics judges every candidate and a miss stops on the spot, and what passes reaches you ranked, with the proof attached. One campaign, step by step: https://www.matterspace.ai/lattice/campaigns ## It learns where to look next. Runtime intelligence that gets sharper with every evaluation. Every round of the search goes where the last one’s proof points. Every evaluation steers the next round. - Guided by evidence: The next round goes where the proof points. - Inverse design: Start from the property. Arrive at the structure. - Diversity on purpose: Different strong answers, not one family of near-copies. - Requirements at runtime: Your limits live inside the search. ## Hundreds of the right candidates, not millions of maybes. Generating millions of structures is easy. Making them is not. MS Lattice holds your requirements inside the search and pushes your preferences as far as they go, so compute goes to contenders, never to a haystack. Your limits act before the compute does. - Requirements, inside the search: Proposals that break your limits are turned away before a structure is built. - Preferences, pushed as far as they go: Nice-to-haves are pushed on top of your must-haves, never at their expense. - A shortlist your lab can make: Ranked candidates, each with its record, instead of a haystack. - Compute spent only on contenders: About a third of a cent per candidate. ## Four modes. One engine. One engine, four kinds of question: from a target with nothing to start from to a benchmark whose answers stay sealed. MS Lattice’s four modes: Open Discovery, Prototype-Guided Optimization, Guided Rediscovery and the Blind Rediscovery Benchmark. What each one takes, what it does and what you get. - Open Discovery: You have a target, not a starting structure. It searches the whole composition space. New materials, from the target alone. - Prototype-Guided Optimization: You have a good material. Make it better. It searches around your structure and tracks every variant against the original. Taking the material you use today further. - Guided Rediscovery: Point it at a known material. Watch it arrive. It lets the targets guide the search and scores how closely every candidate reaches them. 14 of 14 candidates on target in about three minutes, the closest within 0.0008 Å. - Blind Rediscovery Benchmark: Seal the answer. See if it comes back. It searches without the answers, freezes the candidates, then reveals the targets and scores the match. Both sealed ceramics found, down to the arrangement of their atoms. The four modes in depth: https://www.matterspace.ai/lattice/modes ## Validation you can inspect. A score is a claim. MS Lattice gives you the proof behind it: every candidate is judged in stages, and every check it passes stays on its record. Every verdict comes with its reasons. Validation: https://www.matterspace.ai/lattice/validation ## Results: MS Lattice, by the numbers. Sealed targets found blind. The right structure ranked first from a formula alone. Stability and properties called right from first principles. Every candidate rebuilt from its record. - Found blind. Rebuilt to the atom. Two sealed ultra-high-temperature ceramics, found blind, down to the arrangement of their atoms. Four compounds it never saw, CaO, SrF₂, BaTiO₃ and Ni₃Al, rebuilt to within a few hundredths of an ångström. https://www.matterspace.ai/lattice/blind-rediscovery - Inverse design, live. Asked in one sentence for a coating for dry, high-speed machining, 28 GPa, 800 °C in air, under $30 a kilo, MS Lattice put (Ti,Al)N and (Cr,Al)N, the coatings the industry runs on today, in its top five. https://www.matterspace.ai/lattice/inverse-design - First principles, on call. Stability called right four times out of four. Silicon reproduced on five reference values out of five. Elastic constants and phonons on demand. https://www.matterspace.ai/lattice/first-principles - Blind Rediscovery Benchmark: 2 of 2 sealed targets found. In a blind benchmark on ultra-high-temperature ceramics, the two target materials stayed sealed while MS Lattice searched and ranked. When the candidates were frozen and the targets revealed, both had been found, down to the arrangement of their atoms. 64 candidates, in two rounds. About 32 minutes of compute, for under $0.20. Every match level passed, for both targets. Re-run end to end: 9 of 9 checks passed. https://www.matterspace.ai/lattice/results#blind - From the formula alone: 13 of 15 correct structures ranked first. Given nothing but a chemical formula, MS Lattice ranked the correct crystal structure first for 13 of 15 compounds, from rock salt and magnesium oxide to nickel aluminides and magnesium diboride. 13 found, against a bar of 10 set before the run. 9.5 minutes of compute for all fifteen. About 38 seconds per compound. https://www.matterspace.ai/lattice/results#formula - Held-out compounds: 4 of 4 unseen compounds rebuilt. Four compounds were held out of everything MS Lattice starts from: CaO, SrF₂, BaTiO₃ and Ni₃Al. It rebuilt all four to within a few hundredths of an ångström and ranked three of them first. Within 0.016 to 0.037 Å of the reference structures. Three of the four ranked first. https://www.matterspace.ai/lattice/results#held-out - Guided Rediscovery: 14 of 14 candidates on target. Steered towards zirconium diboride, every one of the 14 scored candidates matched it at every level, in a campaign of about three minutes. The closest within 0.0008 Å. Every match level passed by all 14. https://www.matterspace.ai/lattice/results#guided - First-principles stability: 4 of 4 stability verdicts right. MS Lattice’s first-principles stability check called Mg₂Si and MoSi₂ stable and deliberately wrong arrangements of each unstable: four verdicts out of four. Mg₂Si and MoSi₂: 0.165 and 0.284 eV/atom below their competing phases. The wrong arrangements: 0.294 and 0.377 eV/atom above. https://www.matterspace.ai/lattice/results#stability - First-principles properties: 5 of 5 reference values within tolerance. First-principles property calculations reproduced silicon’s lattice constant, its three elastic constants and its highest optical phonon frequency, each within the tolerance set before the run. Lattice constant: 5.4745 Å. Elastic constants C11, C12 and C44: 152.9, 57.0 and 74.0 GPa. Highest optical phonon: 15.18 THz. https://www.matterspace.ai/lattice/results#properties - Reproducibility: 73 of 73 candidates rebuilt from their record. From three production-size campaigns, all 73 audited candidates were rebuilt from the record MS Lattice keeps of each, every property within its declared tolerance. 73 of 73 rebuilt. Every property within its declared tolerance. A full record kept for every candidate. https://www.matterspace.ai/lattice/results#reproduced - From your own agent: 5 min for the whole journey, end to end. Before every release, an AI agent acting as a customer runs the complete journey over MCP: it states a need, gets a priced plan, launches a real cloud campaign and reads back the results. An 8-candidate campaign in 4 to 5 minutes. Under $0.003 of compute. 9,507 automated tests passed on the same release. https://www.matterspace.ai/lattice/results#agent Discovery at about a third of a cent per candidate. ≈ $0.003 of compute per candidate in the blind benchmark. ≈ 30 s of compute per candidate in the same run. ≈ 38 s per compound to rank the right structure from its formula. < $0.0004 per candidate in the agent’s end-to-end campaign. Every result: https://www.matterspace.ai/lattice/results ## The numbers the field reports. Remarkable work. And one clear lesson: the number of candidates was never the product. The answer is. - Microsoft and PNNL: screened 32.6 million candidates on about 1,000 cloud machines, in under 80 hours, to reach 18. - Materials Nexus: analyzed more than 100 million compositions and took three months to design, make and test a magnet. - MS Lattice: 64 candidates. 32 minutes. Under 20 cents. Both sealed targets found. ## The usual pipeline, and MS Lattice. Most AI discovery pipelines generate first and sort later. MS Lattice turns that around. - Your requirements: Held inside the search, before a structure is built - What you get: Hundreds of ranked contenders, each with its record - Preferences: Pushed as far as they go, never at a must-have’s expense - The search: Runtime intelligence that gets sharper with every evaluation - Evidence: Every check on every candidate, rebuildable: 73 of 73 - Compute: About a third of a cent per candidate - AI agents: One MCP connection, 120 tools - Into the real world: Through Norck, on to validation, production and finished parts - Proof: A blind benchmark you can run yourself - Price: Shown before every run How MS Lattice compares: https://www.matterspace.ai/lattice/compare ## Eleven material domains. Batteries, magnets, coatings, catalysts, ceramics. Each domain speaks its own language of targets and limits. MS Lattice speaks all eleven. - Battery cathodes: More energy, less cobalt. Layered NMC oxides, Layered NCA oxides, Spinel oxides, Polyanion frameworks, Sodium-ion layered oxides, Prussian blue analogues, Magnesium-ion spinels. https://www.matterspace.ai/lattice/materials/battery-cathodes - Solid electrolytes: Fast ions, in a stable solid. Garnets, NASICON-type conductors, Sulfides, Halides, Oxide-ion conductors, Proton conductors, Ceramic–polymer composites. https://www.matterspace.ai/lattice/materials/solid-electrolytes - Catalysis: From single active sites to whole surfaces. Single-atom alloys, Alloy surfaces, Oxide-supported catalysts, Acid OER anodes, ORR alloys, CO₂ reduction surfaces. https://www.matterspace.ai/lattice/materials/catalysis - Magnets: More field, fewer rare earths. High-performance NdFeB, Reduced-rare-earth NdFeB, Rare-earth-free MnAl, Hexaferrites, Iron–silicon, Iron–cobalt, Spinel ferrites. https://www.matterspace.ai/lattice/materials/magnets - Photovoltaics: Absorbers for the next solar cells. Halide perovskites, Double perovskites, Chalcohalides, Chalcogenide absorbers. https://www.matterspace.ai/lattice/materials/photovoltaics - Thermoelectrics: Waste heat, put to work. Chalcogenides, Skutterudites, Half-Heuslers, Oxide thermoelectrics. https://www.matterspace.ai/lattice/materials/thermoelectrics - High-entropy alloys: Strength from many elements. Transition-metal alloys, Corrosion-resistant alloys, Lightweight alloys, Refractory alloys. https://www.matterspace.ai/lattice/materials/high-entropy-alloys - Superconductors: Five families, one search. Cuprates, Iron-based pnictides, Bilayer nickelates, MgB₂-type compounds, High-pressure hydrides. https://www.matterspace.ai/lattice/materials/superconductors - Thermal barrier coatings: Hotter engines, protected parts. Stabilized zirconia, Rare-earth zirconates, High-entropy oxides, Environmental barrier coatings. https://www.matterspace.ai/lattice/materials/thermal-barrier-coatings - Hard coatings: Harder, hotter, longer-lasting. Nitrides and carbides, Diborides, MAX phases. https://www.matterspace.ai/lattice/materials/hard-coatings - Ultra-high-temperature ceramics: For the hottest places in flight. Rock-salt carbides, Diboride composites, High-entropy carbides, High-entropy diborides. https://www.matterspace.ai/lattice/materials/ultra-high-temperature-ceramics For ceramic–polymer composites, the solid-electrolyte domain searches for the ceramic filler. Every material domain: https://www.matterspace.ai/lattice/materials ## Your AI agent just became a materials scientist. Your agent states the need, plans the search, runs it for minutes or for weeks, and reads back the proof. MS Lattice is the engine. Your AI agent is the driver. Connect once and your agent can find, compute, verify and prove real materials. Without MS Lattice, an agent can talk about materials. With it, it can discover them. - One connection, 120 tools - Campaigns for minutes or for weeks - Any agent that speaks MCP From your own agent: https://www.matterspace.ai/lattice/agents ## Capabilities - How it works: One MS Lattice campaign, step by step: describe the need, set targets and limits, see the price before you launch, follow the search live, read the ranked candidates and go again from what you kept. https://www.matterspace.ai/lattice/campaigns - Inverse design: Inverse design with MS Lattice: a coating for dry, high-speed machining described in one sentence, and the (Ti,Al)N and (Cr,Al)N coatings the industry runs on today came back in the top five. https://www.matterspace.ai/lattice/inverse-design - The engine: The MS Lattice engine: runtime intelligence that learns where to look next, your limits inside every step, the right estimate for every property and every candidate on record. https://www.matterspace.ai/lattice/engine - What it costs: What MS Lattice costs: about a third of a cent of compute per candidate, a blind rediscovery run for under 20 cents, and the price of every plan shown before it launches. https://www.matterspace.ai/lattice/economics - Blind rediscovery: The Blind Rediscovery Benchmark: two sealed ultra-high-temperature ceramics found blind by MS Lattice, among 64 candidates, in 32 minutes, for under 20 cents. https://www.matterspace.ai/lattice/blind-rediscovery - First principles: First-principles verification in MS Lattice: stability against rival phases, elastic constants and phonons on demand. 4 of 4 stability verdicts right; silicon reproduced on 5 of 5 reference values. https://www.matterspace.ai/lattice/first-principles - Validation: How MS Lattice proves every candidate: evaluated in stages, every check on record, uncertainty with every estimate, stability against every rival phase, novelty against known crystals and checks on demand. https://www.matterspace.ai/lattice/validation - Agents and MCP: Connect your AI agent to MS Lattice over MCP: one connection, 120 tools, the whole chain from need to proof. The same engine through the REST API, the Python SDK, the command line and the web dashboard. https://www.matterspace.ai/lattice/agents ## Two ways to work. License MS Lattice for your own team and agents, or let Norck take your target all the way to a finished part. - License MS Lattice: Every domain, every mode, from your own AI agent over MCP, the REST API, the Python SDK or the command line. - Discovery with Norck: Norck runs MS Lattice, then synthesizes, validates and manufactures the winner, all the way to your finished part. Two ways to work with MatterSpace: https://www.matterspace.ai/services ## From the search to your lab. MS Lattice does the search and the proof. Your lab starts from the shortlist. - Ready for the lab: A ranked shortlist, every candidate with its record. - Exported as structures: A standard structure file for every candidate. - Proven blind: Both sealed targets found, and four compounds it had never seen rebuilt, without being shown. - On to manufacturing: Through Norck, a discovery becomes a finished part. ## The first of our discovery engines. MS Lattice is the first of MatterSpace’s discovery engines. More are on the way, for every problem in science and engineering whose space of possibilities is too large to search one try at a time. Bring us a materials target: https://www.matterspace.ai/lattice/contact Trademarks of Machine Autonomy, Inc.: MatterSpace, MS Lattice and the company slogan “Advancing Bits to Master Atoms”. Website content © 2026 Machine Autonomy, Inc.