LUNAROPS · OPERATIONAL UPLINK 100% UPTIME 1,247d POSTS 893 JEFF.MOON@LUNAROPS.DEV UTC --:--:--

Software Patents for Working Engineers

16 min read Law & Standards
software-patentsintellectual-propertypatent-lawpatent-trollsopen-sourcelaw-and-standards
Contents

Most engineers read patents wrong, because patents are written to be read wrong. The title is marketing. The abstract is marketing. The drawings are decoration. The forty paragraphs describing “one embodiment of the present invention” are deliberately broad smoke. The only part of a patent with legal force is the numbered list of claims at the very end — and claims are read under rules so specific that two patents with identical titles can have completely different scope. You will interact with this system whether you want to or not: your employment contract almost certainly contains an invention assignment clause, your employer may dangle filing bonuses at you, your side project could receive a demand letter, and the prior art that kills a bad patent might be a blog post you wrote. This is the working engineer’s guide to a legal system that regulates your output.

A Patent Is a Boolean Expression

A US utility patent gives its owner the right to exclude others from making, using, or selling the claimed invention for 20 years from the filing date. Not the right to practice it — only the right to stop others. The document has a fixed anatomy: title, abstract, drawings, a long narrative section called the specification, and the claims. Infringement analysis ignores everything except the claims.

Each claim is a single sentence (sometimes hundreds of words long) listing limitations. A product infringes a claim only if it practices every limitation. Miss one and that claim does not read on your product. This is the all-elements rule, and it means infringement is a boolean AND:

infringes(product, claim) :=
    AND over every limitation L in claim:
        product practices L

Claim 1 of US 5,960,411 ("one-click"), paraphrased:
  L1: client displays item information
  L2: in response to ONLY A SINGLE ACTION being performed,
      a request to order the item is sent to a server
  L3: server uses a pre-stored identifier to retrieve
      purchaser account data
  L4: the order is fulfilled WITHOUT using a shopping
      cart ordering model

A checkout that requires two clicks fails L2. Done. This is
why Barnes & Noble's "fix" was literally adding a second
click, and why design-arounds are a real legal strategy.

Claims come in two flavors. Independent claims stand alone and define the broadest scope. Dependent claims reference a parent (“The method of claim 1, wherein…”) and narrow it; they exist as fallback positions, because if claim 1 is invalidated in litigation, narrower claim 5 might survive. When you skim a patent to assess risk, read the independent claims first — there are usually only two or three — and ask whether every limitation maps to something you actually do.

Two more mechanics matter. First, the doctrine of equivalents can catch a product that avoids a limitation’s literal text but does substantially the same thing in substantially the same way — so a cute rename does not save you, though a genuine architectural difference does. Second, claim terms mean what the specification says they mean, not what a dictionary says. Litigation over a single word’s construction (a Markman hearing) routinely decides entire cases.

The Eligibility Rollercoaster: Benson to Alice

Whether software is patentable at all has whipsawed for fifty years, and the case law reads like a control system hunting for a setpoint it never finds.

Case Year Holding Practical effect
Gottschalk v. Benson 1972 BCD-to-binary conversion algorithm is an unpatentable abstract idea Pure algorithms excluded
Parker v. Flook 1978 Updating an alarm limit with a formula is ineligible “Post-solution activity” doesn’t save a formula
Diamond v. Diehr 1981 Rubber-curing process using the Arrhenius equation IS eligible Software inside an industrial process is fine
State Street Bank 1998 “Useful, concrete and tangible result” suffices (Fed. Cir.) Floodgates open; business-method patents boom
Bilski v. Kappos 2010 Hedging risk is abstract; machine-or-transformation is a clue, not the test First brake on the boom
Mayo v. Prometheus 2012 Two-step framework for laws of nature The scaffolding Alice will reuse
Alice v. CLS Bank 2014 Generic computer implementation of an abstract idea is ineligible Thousands of software patents die

Alice Corp. v. CLS Bank International (2014) is the one to actually understand. Alice owned patents on computerized escrow — using a third-party intermediary to settle financial transactions, implemented on a generic computer. The Supreme Court unanimously invalidated them and, in doing so, fixed the two-step test from Mayo as the gate every software patent must pass under 35 U.S.C. §101:

            +--------------------------------------+
            | §101 threshold: process, machine,    |
            | manufacture, or composition of       |
            | matter?                              |
            +-------------------+------------------+
                                | yes
            +-------------------v------------------+
  STEP 1    | Is the claim "directed to" an        |  no
            | abstract idea, law of nature, or     +-----> ELIGIBLE
            | natural phenomenon?                  |
            +-------------------+------------------+
                                | yes
            +-------------------v------------------+
  STEP 2    | Does it add an "inventive concept" — |  yes
            | significantly more than the abstract +-----> ELIGIBLE
            | idea performed on a generic          |
            | computer?                            |
            +-------------------+------------------+
                                | no
                                v
                    INELIGIBLE under §101

The Court’s key sentence: “merely requiring generic computer implementation fails to transform that abstract idea into a patent-eligible invention.” Take a centuries-old practice — escrow, hedging, advertising-as-currency — append “on a computer,” and you have nothing. In the eighteen months after Alice, district courts and the Federal Circuit invalidated patents under §101 at rates above 70%, and the USPTO’s allowance rate for business-method art units fell off a cliff.

What survives? The Federal Circuit has spent a decade drawing the line case by case. DDR Holdings (2014) survived because the claims solved a problem “necessarily rooted in computer technology” (retaining website visitors across outbound links). Enfish (2016) survived at step 1 because a self-referential database table was an improvement to computer functionality itself, not a business practice using a computer. McRO (2016) survived because specific rules for automated lip-sync animation were not just “animate it on a computer.” Berkheimer (2018) added that whether something is “well-understood, routine, and conventional” is a question of fact — making early dismissal harder.

The honest summary for an engineer: a genuine technical improvement — a measurably better data structure, protocol, compression scheme, memory layout, rendering pipeline — is still patentable. “Doing a known thing, but on a computer / on the internet / with a database / with machine learning” is not, though plenty of such patents still issue and must be killed expensively later. Note also that this is a US story: the European Patent Convention nominally excludes “programs for computers as such,” but the EPO grants computer-implemented inventions that produce a “further technical effect” — in practice the two systems converge more than their statutes suggest.

Congress has repeatedly tried to overwrite Alice — the Tillis–Coons Patent Eligibility Restoration Act has been reintroduced several times, most recently in 2025 — but as of this writing none of these bills has reached a floor vote. Plan for Alice to remain the law.

Trolls: The Economics of Asymmetric Litigation

A non-practicing entity (NPE) — a company whose only product is litigation — exists because of an arbitrage: defending a patent suit through trial costs $1M–$4M+ (AIPLA’s economic surveys have shown this for years), while a settlement demand of $50k–$300k is rationally payable even when the patent is garbage. The NPE buys patents from defunct startups or shrinking companies for cents on the dollar, sends hundreds of demand letters, and profits on volume. Operating companies sue each other too, but in high tech, NPEs have accounted for the majority of patent litigation in most recent years (Unified Patents’ annual reports put it around 60%).

Geography was part of the model. For two decades, the Eastern District of Texas — Marshall, population ~24,000 — heard more patent cases than any other US district, thanks to plaintiff-friendly local rules and venue law that let you sue a national company anywhere it sold products. TC Heartland v. Kraft Foods (2017) ended that, restricting venue to where the defendant is incorporated or has a regular and established place of business. The caseload promptly migrated — much of it to the Western District of Texas in Waco — but the structural correction was real.

The other structural correction was the America Invents Act (2011), which created inter partes review: a proceeding before the Patent Trial and Appeal Board where anyone can challenge a patent’s validity using prior art, for low six figures all-in instead of millions, decided by technically trained judges in about 18 months. Instituted IPRs invalidate at least some challenged claims roughly 80% of the time — a former Federal Circuit chief judge angrily called the PTAB “death squads killing property rights,” which tells you it works. If an NPE’s patent has already lost an IPR, its demand letter is waste paper; checking PTAB history is step one of any triage.

The AIA also switched the US from first-to-invent to first-to-file (March 2013), aligning it with the rest of the world. Your lab notebook no longer wins races; your filing date does — with one exception that matters for defense, covered next.

Defensive Strategies: Making Yourself Unpatentable-Against

You do not need to file patents to defend yourself against them. The cheapest and most underused defensive tool is publication.

Strategy Cost What it does The catch
Defensive publication (TDCommons, IP.com) Free–$400 Creates dated prior art; nobody can later patent what you published Protects nothing offensively; you give the idea away
Your own blog/repo/conference talk Free Same prior-art effect if dated and public Must be findable; examiners rarely search blogs, but IPR petitioners do
Filing your own patents $15k–$30k+ each with counsel Deterrent counter-arsenal; trade value Useless against NPEs — they have no products to countersue
LOT Network membership Free under $25M revenue If any member’s patent is sold to an NPE, all members get an automatic license Doesn’t stop operating-company suits between members
Open Invention Network Free Royalty-free cross-license covering the “Linux System” (thousands of packages, incl. Android, Kubernetes) Scope limited to the Linux System definition
Inter partes review ~$100k–$500k Kills bad patents at the source Only worth it once you’re actually targeted

Prior art is anything publicly available before a patent’s effective filing date: papers, manuals, source code, mailing list posts, products. A patent claim that reads on prior art is invalid — the system’s one self-correcting mechanism. Defensive publication weaponizes this: Google runs TDCommons as a free venue where companies dump inventions they don’t want to patent but don’t want patented against them. Publishing a detailed technical writeup of your clever scheme costs you an afternoon and permanently removes it from everyone’s patentable space. If you already write up your projects publicly, you are doing this by accident — one more reason building in public pays off beyond marketing. (One nuance: under the AIA, your own publication starts a one-year clock during which you can still file in the US; everyone else is blocked immediately.)

The collective-defense structures are worth knowing by name. The LOT Network (License on Transfer, founded 2014 by Google, Canon, and Red Hat) is an elegant hack: thousands of member companies agree that the moment any member’s patent is transferred to an NPE, every other member automatically receives a free license to it. Members retain full use of their patents against each other — it surgically removes only the troll exit. Membership is free below $25M revenue, which means there is little excuse for a funded startup not to join. The Open Invention Network (founded 2005) is a royalty-free cross-license among 3,000+ participants covering core open-source infrastructure; its biggest symbolic win was Microsoft joining in 2018, bringing 60,000 patents — including the exFAT patents it had spent a decade taxing Android vendors with — inside the non-aggression pact, which is why exFAT support landed in mainline Linux in 2019.

Patents at Your Day Job

The paperwork you signed on day one almost certainly assigned your future inventions to your employer. Read it anyway, because the scope varies and the law limits it. California Labor Code §2870 (and similar statutes in roughly ten other states) voids assignment of inventions you develop entirely on your own time, without employer equipment or trade secrets — unless the invention relates to the employer’s business or anticipated R&D, or results from your work for them. That carve-out’s carve-out is the trap: at a large company whose business “relates to” most of computing, your weekend project may still be theirs on paper. If you have a serious side project, get the exclusion list right when you sign, not after it ships — the same diligence that applies to side projects that ship and to reviewing terms during a career or job change. Watch for trailer (holdover) clauses too, which claim inventions conceived for months after you leave.

Inside the company, the pipeline runs on invention disclosure forms. You write a one-pager describing what you built and why it’s novel; a patent committee (attorneys plus senior engineers) scores it; winners go to outside counsel who drafts claims you will barely recognize as your work. Typical incentives are $1k–$5k per filing plus a plaque. Participate or don’t — but understand the two honest framings. The cynical one: most corporate filings are portfolio chaff for cross-licensing leverage, and your elegant idea becomes claim 14 of a patent asserted in a dispute you’ll never hear about. The pragmatic one: patents on your resume are legible evidence of invention, the bonus is real money, and a defensive portfolio genuinely does deter operating-company suits against your employer — which protects your job.

One counterintuitive corporate policy deserves explanation: many employers instruct engineers not to read patents. This is willfulness management. Infringement you knew about can be “willful,” exposing the company to up to treble damages under 35 U.S.C. §284 (Halo v. Pulse, 2016, made enhancement easier to get). An engineer’s Slack message saying “this looks like patent X” is discoverable. The policy feels anti-intellectual; it is rational risk management under bad incentive design. Related discipline applies when studying competitor products: keep provenance clean, the same way clean-room practice governs reverse engineering with Ghidra.

If a Demand Letter Arrives

For a small company or solo developer, the playbook is short:

  1. Do not reply, and do not panic. The letter is engineered to scare you into a fast settlement. Most demand-letter campaigns never escalate to filed suits — suing costs the troll real money.
  2. Do not ignore it forever either. Knowledge is now established; continued infringement, if real, can be willful. Calibrate, don’t freeze.
  3. Triage the patent. Look it up on Google Patents. Check PTAB for IPRs (dead or wounded claims are common). Search the asserter on RPX, Unified Patents, or Lex Machina data — serial NPEs have public track records showing what they actually accept.
  4. Map the independent claims against what you ship. All-elements rule: find one limitation you don’t practice and document it.
  5. Get a patent litigator for one consultation before any substantive response. An opinion-of-counsel letter also rebuts later willfulness claims.
  6. Check your dependencies’ indemnities. Cloud providers and many commercial vendors indemnify customers against patent claims arising from their services; the letter may be someone else’s problem contractually.
  7. Tell the EFF if the demand is abusive — their patent project has repeatedly crowdsourced prior art that killed troll patents.

The cautionary tale for over-fighting is NTP v. RIM: RIM litigated for years, the patents were being shredded in reexamination, and RIM still paid $612.5M in 2006 because an injunction threatened to shut down BlackBerry service before the reexams finished. The cautionary tale for under-fighting is every company that quietly paid a five-figure settlement on a patent that an afternoon of prior-art searching would have embarrassed.

The Greatest Hits

The famous fights are worth knowing because they shaped both the law and the industry’s defensive culture. Amazon’s one-click patent (US 5,960,411, granted 1999) produced a holiday-season injunction against Barnes & Noble, a public apology tour in which even Bezos called for shorter software patent terms, a community-funded reexamination that narrowed the claims, and seventeen years of Apple paying license fees for the Apple Store’s one-click checkout before expiry in 2017. Unisys and the GIF: the LZW compression patent, asserted starting in 1994 against a format everyone thought was free, triggered “Burn All GIFs” day and directly caused the creation of PNG — the clearest case of a patent producing a better open replacement. Fraunhofer’s MP3 patents kept the LAME encoder distributed as source-only for years and kept MP3 support out of Fedora until the last patents expired in 2017. i4i v. Microsoft (2011) put a $290M price on custom XML in Word and forced a shipping feature’s removal. Microsoft’s Android tax — FAT long-filename and exFAT patents licensed to nearly every Android OEM, reportedly approaching a billion dollars a year from Samsung alone at peak — shows patents as a pure revenue layer on a competitor’s product, and its wind-down via OIN membership shows how the collective-defense structures actually get used.

And the biggest one was not a patent case at all. Oracle v. Google began in 2010 with seven patents and a copyright claim over 37 Java API packages; the patent claims died at trial, and the copyright fight over whether API declarations — the java.lang.Math.max() signatures themselves — are protectable ran eleven years to the Supreme Court. The 2021 decision (6–2) held Google’s use was fair use while pointedly declining to say whether APIs are copyrightable at all. Every API you reimplement for compatibility lives under that unanswered question, which is reason enough to treat interface design as the durable public contract it is — see the API design principles post.

Verdict

The software patent system after Alice is better than its reputation but worse than it should be. Genuine technical inventions remain patentable, garbage still issues, and the correction mechanisms — IPR, prior art, venue reform — work but only when someone pays to invoke them. As a working engineer, your rational posture is mostly defensive: read claims, not titles, when assessing risk; publish technical writeups early because dated public disclosure is free insurance; get your side-project IP carve-outs in writing on day one, not at dispute time; push your employer toward LOT and OIN if it isn’t already a member; and treat demand letters as a triage problem with a known playbook rather than an existential threat. File patents through your employer’s program if the bonus and the resume line are worth an afternoon of paperwork to you — just don’t confuse claim 14 of a cross-licensing asset with ownership of your idea. The system regulates your output whether you engage with it or not. Engage with the cheap, high-leverage parts.

Sources

Comments