Patent Eligible Subject Matter

Section 101 sets the threshold for what can be patented in the United States: THE STATUTORY TEXT: '§ 101. Inventions patentable. Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title'; THE FOUR STATUTORY CATEGORIES: PROCESS: a series of steps or acts; includes methods of treatment, methods of manufacture, software processes, business methods; MACHINE: a concrete device or combination of devices; includes computers, robots, sensors, chemical reactors; MANUFACTURE: an article produced from raw materials; includes fabricated objects, non-naturally occurring materials; COMPOSITION OF MATTER: chemical compositions; mixtures; compounds; biological compositions (subject to judicial exceptions); DIAMOND v. CHAKRABARTY (S.Ct. 1980): a bacterial strain genetically engineered to digest oil spills was held patent-eligible; the Court said Congress intended patentability to 'include anything under the sun that is made by man'; the scope of § 101 was intended to be BROAD; excluded: naturally occurring phenomena; mathematical laws; abstract ideas; THREE REQUIREMENTS FOR PATENTABILITY: § 101 is a THRESHOLD test (is this the type of thing that can be patented?); additional requirements must also be met: § 102 (novelty — not anticipated); § 103 (non-obviousness); § 112 (specification requirements); THE JUDICIAL EXCEPTIONS: despite the broad language, courts created three exceptions: (1) laws of nature — naturally occurring relationships (E=mc²; the correlation in Mayo; Prometheus); (2) natural phenomena — products of nature (isolated DNA in Myriad; the cell-free fetal DNA in Ariosa); (3) abstract ideas — mathematical concepts, mental processes, methods of organizing human activity (Bilski; Alice); NOTE ON UTILITY: § 101 also requires the invention to be 'useful' — the utility requirement; this is rarely an issue except in very early-stage research compounds where utility has not been established.

Each statutory category covers different types of inventions in modern technology: PROCESS CLAIMS — MOST COMMON IN SOFTWARE AND BIOTECH: a sequence of steps performed to achieve a result; software: a method of encrypting data using specific steps; a method of training a neural network; biotech: a method of producing a recombinant protein; a method of treating a disease; the process category is intentionally broad — it includes 'anything done or effected by a series of acts'; MACHINE CLAIMS — MOST COMMON IN HARDWARE: a concrete apparatus; computer hardware systems; smartphones; sensors and IoT devices; specific computer architectures; manufacturing equipment; a machine claim describes the physical components and their interconnections; machines are generally considered safer from § 101 challenges than method claims for the same invention; MANUFACTURE CLAIMS: products produced by human effort; pharmaceuticals (compound claims); electronic components; medical devices; semiconductor chips; manufactured products are generally eligible unless they run into the natural phenomenon exception (Myriad: isolated DNA = natural phenomenon; cDNA = manufacture); COMPOSITION OF MATTER: all compounds and mixtures regardless of whether their ingredients are naturally occurring; patented drugs are typically compositions of matter (specific chemical compound claim); polymers; alloys; biological compositions (vaccines; antibodies — subject to Myriad); PRACTICAL STRATEGY — MULTIPLE CLAIM TYPES: skilled patent practitioners draft claims in multiple statutory categories for the same invention; a software invention may be claimed as: a method (process); a system comprising a processor + memory + software (machine); a computer-readable medium (non-transitory CRM — manufacture); this 'three-track' approach provides multiple avenues for enforcement even if one type faces § 101 challenges.

The current Alice/Mayo framework was consolidated through a series of Supreme Court decisions: STEP 1 — DIRECTED TO A JUDICIAL EXCEPTION?: courts ask whether the claim is 'directed to' a law of nature, natural phenomenon, or abstract idea; the claim is analyzed at the level of its 'focus' or 'character as a whole'; this is not about whether the claim merely mentions a law of nature or abstract concept — it is about whether the claim's central innovation is the exception; WHAT 'DIRECTED TO' MEANS: a claim may 'involve' an abstract idea or law of nature without being 'directed to' it; a computer claim can use mathematical operations (directed to computer improvement, not to the math); a treatment claim can rely on a natural correlation (directed to treatment, not to the natural correlation); STEP 2 — SIGNIFICANTLY MORE: if a claim IS directed to a judicial exception, look for additional elements that 'transform' the claim into patent-eligible subject matter; the claim must add 'significantly more' than the abstract idea, law, or phenomenon itself; STEP 2 FAILS (ineligible): generic computer implementation; well-understood, routine, conventional activity in the relevant field; mere instructions to 'apply it'; post-solution activities (routine data gathering + compute + apply to conventional process); STEP 2 SURVIVES (eligible): specific technical improvements to computer technology; non-conventional ordered combination of steps; specific machines with non-conventional configurations; integration of the judicial exception into a larger non-conventional process; 2019 USPTO REVISED GUIDANCE: adds Step 2A Prong 2 (practical application integration) — a claim integrating a judicial exception into a specific practical application may be eligible even without 'significantly more' per traditional Step 2; courts are NOT bound by USPTO guidance but it governs examination.

AI and machine learning patents present unique § 101 challenges: MACHINE LEARNING ARCHITECTURE CLAIMS: claims to specific neural network architectures (specific connections; specific training algorithms; specific activation functions) that improve computer performance may be eligible; the key is whether the architecture itself is a technical innovation or merely an application of known neural network techniques to a new problem; 'train a neural network on X to predict Y' without specific technical improvements = likely abstract; APPLICATION-SPECIFIC AI CLAIMS: claims to specific AI applications (facial recognition; autonomous vehicle navigation; drug discovery prediction) face scrutiny as to whether the claim is anchored to the AI's specific technical mechanism or just to the result; 'a neural network that identifies cancer in medical images' — if the claim just covers this output without specifying technical innovation, it risks § 101 rejection; TRAINING DATA AND METHODS: claims to novel training data curation methods or novel training objectives (loss functions) may be eligible if they represent specific technical innovations; 'selecting training data using specific non-conventional criteria' + specific architecture + specific training approach = potentially eligible; TRANSFORMER ARCHITECTURE: claims to specific improvements to transformer architecture (attention mechanisms with specific novel configurations) may be eligible as technical improvements to computer hardware/software; LARGE LANGUAGE MODEL PATENTS: claims to specific improvements in LLM training (specific quantization approaches; specific fine-tuning methods with specific technical advantages) may be eligible; claims to 'an LLM that answers questions' = likely abstract; USPTO AI GUIDANCE: the USPTO has issued guidance on AI patent eligibility; claims should be directed to specific technical improvements (memory efficiency; computational speed; accuracy) demonstrated through specific technical mechanisms; PRACTICAL APPROACH: draft AI patent claims that are anchored to specific technical architectures and that describe specific technical advantages; include working examples and performance benchmarks in the specification.

Section 101 has been one of the most controversial areas of patent law in recent decades: HISTORY: § 101 as currently written derives from the Patent Act of 1952; the language 'anything under the sun made by man' (Diamond v. Chakrabarty) reflected the expansive view Congress intended; the judicial exceptions were created by the Supreme Court, not Congress; BILSKI v. KAPPOS (S.Ct. 2010): business method patent for commodities hedging held ineligible; rejected the Federal Circuit's 'machine-or-transformation test' as the SOLE test for patent eligibility (though it remains a useful clue); opened the door to broader § 101 challenges; MAYO (2012) + ALICE (2014) + MYRIAD (2013): the three-year period that reshaped § 101 dramatically; thousands of previously issued patents suddenly faced validity challenges; INDUSTRY REACTIONS: software industry (Google, Amazon, Facebook): generally supportive of broad Alice interpretation — reduces threat from NPE software patent assertions; biotech and pharma: strongly opposed to Mayo/Alice — valuable diagnostics and personalized medicine patents invalidated; universities: concerned about loss of diagnostic patent value; traditional manufacturers: supportive of § 101 reform to restore eligibility; PROPOSED LEGISLATION: multiple bills to reform § 101 have been introduced in Congress; the 2019 Tillis-Coons proposal would have eliminated judicial exceptions; in 2023-2024, the Patent Eligibility Restoration Act (PERA) was proposed — would eliminate judicial exceptions and replace them with a statutory list of excluded subject matter; none of these proposals have passed; CURRENT REFORM EFFORTS: the debate continues; critics of current law argue: Alice/Mayo are too vague; many valuable innovations (diagnostics, AI, software) are now unprotectable; international competitors (China, EU) have more permissive § 101 equivalents; defenders argue: Alice/Mayo prevent patent thickets; protect against overly broad patent claims that would harm competition.