How software patents are like armed career criminals
(They are mostly not alike.)
It was a pleasure to collaborate with my colleagues at Jenner & Block, Matt Hellman and Yoni Marshall, on an amicus brief in Amgen Inc. v. Sanofi filed earlier today on behalf of the Association for Accessible Medicines.
Amgen concerns the Patent Act’s “enablement” requirement—that is, the requirement that the “specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same.” Amgen developed an antibody that binds to and inhibits a protein known as PCSK9. It then managed to obtain a patent purporting to cover all antibodies that bind to and inhibit PCSK9—of which there are millions. It sued Sanofi for infringement, based on Sanofi’s independent development of an antibody that was both different and better than the antibody that Amgen had developed.
Thanks for reading Adam's Legal Newsletter! Subscribe for free to receive new posts and support my work.
The Federal Circuit held that Amgen’s patent was not enabled because the specification doesn’t explain how to make and use Sanofi’s antibody and the millions of other antibodies it purportedly covers. For the reasons stated in our amicus brief, I believe the Federal Circuit’s decision is persuasive and correct, and I hope the Supreme Court affirms.
Amgen is a pharma case. I am a tech person at heart, so I’d like to discuss the implications of Amgen and the enablement doctrine on the validity of software patents. In this post I will argue:
In recent years, the Federal Circuit has attempted to rein in overbroad software patents by declaring them ineligible to be patented under 35 U.S.C. § 101.
Overbroad software patents are a blight. But experience has shown that applying § 101 to overbroad software patents is impossible. The legal standard is so formless and subjective that it is incapable of being applied in a non-arbitrary way.
In Johnson v. United States, 576 U.S. 591 (2015), the Supreme Court struck down the residual clause of the Armed Career Criminal Act on the ground that it was so unpredictable and arbitrary that applying it violated the Due Process Clause. As currently being applied, § 101 is a lot like the residual clause. The problem isn’t that § 101 is ambiguous or that there are hard or borderline cases. The problem is that, as with ACCA’s residual clause, courts attempting to apply § 101 do not understand what question they are supposed to answer, much less what they are supposed to do to answer it.
The enablement doctrine is the beginning of a way out. It at least provides some semblance of a legal standard, as opposed to grappling in the pitch darkness of § 101.
E=mc2, coelacanths, and existentialism …
… are some things you can’t patent.
Under 35 U.S.C. § 101, one can obtain a patent on “any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof.” That sounds like pretty much everything. But for 150 years, the Supreme Court has held that there is an implied exception to § 101: “laws of nature, natural phenomena, and abstract ideas” are not patentable.
Why? To get a patent, you have to invent new technology. Laws of nature, natural phenomena, and abstract ideas aren’t technology.
Suppose a patentee discovers (e.g.) a new law of nature that, when applied, is useful. Can the patentee get a patent on the use of the law of nature, as opposed to the law of nature itself? No, held the Supreme Court in 2012. Patenting the use of the law is, as a practical matter, the same thing as patenting the law.
The concepts of “laws of nature” and “natural phenomena” seem fairly straightforward, although they generate close cases at the margin. But what are “abstract ideas”?
No concept is more of an abstract idea than the concept of an abstract idea. But there are a few things that people seem to agree are abstract ideas.
Mathematical formulas and algorithms are abstract ideas. Suppose a mathematician develops an algorithm to factor large integers quickly.1 No one has ever developed such an algorithm, but no one has ever proven there isn't one, so you never know. Such an algorithm would end modern cryptography as we know it, but its inventor wouldn't be allowed to patent it.
Non-technological intellectual concepts are also abstract ideas. For example, Bilski v. Kappos, 561 U.S. 593 (2010), holds that you can’t get a patent on the use of hedging in energy transactions, even if you reduce that idea to a mathematical formula.
What about if you try to implement an abstract idea on a generic computer? This topic was addressed in Alice Corporation v. CLS Bank International, 573 U.S. 208 (2014).
The patent in Alice described the exchange of financial obligations between two parties using a third-party intermediary to mitigate settlement risk. This is a classic non-technological intellectual concept. It’s just … two people using a trusted third party to facilitate a deal. But the twist was, the claims said the financial intermediation would occur using computers. And, indeed, computers make third-party intermediation much faster and easier. So the patentee argued that it had invented technology. A computer is a type of machine, so a computer that does intermediation is a new piece of technology, or so the argument went.
The Supreme Court unanimously rejected this argument. In the Supreme Court’s view, the invention was the intellectual concept; merely using a generic computer to implement it didn’t transform it into a patentable invention.
Seems straightforward enough, but then things started getting weird.
The abstractness of RAM
Bilski and Alice hold that non-technological, intellectual concepts are unpatentable, even if codified into a mathematical formula or implemented on a generic computer.
In recent years, however, the Federal Circuit has attempted to apply § 101 to a different type of patent: overbroad software patents. The problem in these cases isn’t that the purported invention is a non-technological intellectual concept. The problem is that the purported invention is an idea for implementing software, rather than the actual implementation of software.
There are many examples of this type of case, but Visual Memory LLC v. Nvidia Corp., 867 F.3d 1253 (Fed. Cir. 2017), is a particularly good one.
Some background first. When you buy a computer, it comes with many types of memory. One type of memory is the cache. The cache is really fast, but also really expensive, so there isn’t much of it. There’s also RAM, which is slower but cheaper than the cache. Finally, there’s storage memory—i.e., the hard drive—which is slower but cheaper than RAM. Together, these different types of memories are called a memory system.
For efficiency’s sake, the cache contains the data that the processor uses most frequently. However, which data is used most frequently depends on the processor.
According to the patent in Visual Memory, prior art memory systems were designed to be compatible with particular processors. If the processor used particular types of data a lot, the associated prior art memory system would have a cache that stored that type of data efficiently. But if you tried to use the prior art memory system on a different processor, the memory system would be inefficient because the different processor might use other types of data more often.
The innovation in Visual Memory was: programmable memory. Suppose you wanted to use a memory system with a new processor. If you could simply reprogram the memory system to be tailored to the new processor, this problem would be solved!
The inventors persuaded the Patent Office to give them a patent on programmable memory. Here’s what the broadest claim said:
1. A computer memory system connectable to a processor and having one or more programmable operational characteristics, said characteristics being defined through configuration by said computer based on the type of said processor, wherein said system is connectable to said processor by a bus, said system comprising:
a main memory connected to said bus; and
a cache connected to said bus;
wherein a programmable operational characteristic of said system determines a type of data stored by said cache.
A “bus” is just a communication system that transfers data from one computer component to another. So this claim covers a memory system, connected to a processor, that can be programmed to store particular types of data. Any memory system meeting this description, regardless of how it is programmed, infringes.
Is this patentable?
My first reaction is, no, of course it shouldn’t be patentable. But it is hard to put a finger on why.
One reason is that it seems obvious. Prior to this “invention,” memories were inefficient when used with new processors. Is it really so inventive to say, “well, why don’t you just reprogram the memory and make it better!” This seems like something an elementary school student might propose.
But a court cannot wipe out the patent based on a mere instinct that the patent is obvious. To establish obviousness, accused infringers have to locate prior art, and then argue that the patent is obvious in light of that prior art. Prior art sometimes isn’t so easy to find, particularly for older inventions.
Should judges should have the power to declare a patent obvious without prior art, maybe even at the motion to dismiss stage? I don’t think so. How is the judge supposed to know whether the patent is obvious? Or should the judge only invalidate patents that are obviously obvious? This is not a recipe for consistent outcomes.
Also, obviousness doesn’t capture the intuitive problem here. The problem is that this isn’t an … invention. It’s an idea for an invention, but the invention hasn’t been invented yet. Intuitively, a patentee should be someone who has an idea, develops a protoype, tinkers with it, figures out how to make it work, and obtains a patent based on the fruit of her labors. Here, the claim consists of … “let’s program memory to make it work better!”
But is this claim an “abstract idea,” akin to hedging or using financial intermediaries, that is invalid as a matter of law? That’s a much harder question.
As regularly occurs in 101 cases, the Federal Circuit panel divided 2-1. The majority held that the claims were patent-eligible because they “are directed to a technological improvement: an enhanced computer memory system.” The dissent said that no, “the ‘740 claims are directed to the abstract idea of categorical data storage.”
Who was right?
Is possessing a sawed-off shotgun riskier than burgling a house?
Let’s take a detour to the statute that the Supreme Court hates the most: the Armed Career Criminal Act (otherwise known as ACCA).
It’s a federal crime for a convicted felon to possess a firearm. Usually, the statutory maximum sentence is 10 years. But that 10 year maximum transforms into a 15 year minimum if the defendant is has previously committed three violent felonies. “Violent felony” is defined, in relevant part, as a felony that:
“is burglary, arson, or extortion, involves use of explosives, or otherwise involves conduct that presents a serious potential risk of physical injury to another.”
This language has proven challenging to interpret, to put it mildly. As to the first part of the definition—the enumerated offenses of burglary, arson, extortion, and the use of explosives—the Supreme Court has issued a series of bitterly divided decisions culminating in Justice Alito comparing the Court to Sabine Moreau, a Belgian driver who took a wrong turn and accidentally drove to Zagreb, Croatia. Ms. Moreau eventually figured out her error, but according to Justice Alito, “the Court, unlike Ms. Moreau, is determined to stay the course and continue on, traveling even further away from the intended destination.” Justice Alito was wistful: “Who knows when, if ever, the Court will call home.”
The first part of the definition is the easier part to interpret. The second part—the so-called “residual clause,” covering any felony that “otherwise involves conduct that presents a serious potential risk of physical injury to another”—is so vague that the Supreme Court, in Johnson v. United States, 576 U.S. 591 (2015), took the bold step of declaring it facially unconstitutional. In a later case, Welch v. United States, 578 U.S. 120 (2016), held that Johnson applies retroactively, meaning that all violent felons given enhancements under the residual clause, past and present, have had those enhancements wiped out.
Why did the Court do this? The Court had heard a series of cases involving whether particular crimes were violent felonies under the residual clause. These cases had involved attempted burglary, drunk driving, failure to report to prison, and vehicular flight. In every case, the Court applied a different, ad hoc test to assess whether the crime “otherwise involves conduct that presents a serious potential risk of physical injury to another.” Johnson involved possession of a sawed-off shotgun. No one had any idea what test to apply, or how to apply it. The Court had had enough, and declared the residual clause unconstitutionally vague.
Lots of criminal statutes are ambiguous. Almost always, the Court interprets them (sometimes narrowly) rather than striking them down. But the residual clause of ACCA had three characteristics that made it special.
No one knew how to determine how risky the unenumerated crime is. How risky is possessing a sawed-off shotgun? Seems risky, but does it present “serious potential risk,” in the words of the statute? Does “potential risk” mean something different from “risk”? Is possessing a sawed-off shotgun riskier (or potentially riskier) than burglary or extortion? How risky are burglary or extortion anyway? Are they riskier than failing to report to prison or drunk driving?
No one knew what test to use. The residual clause covered the enumerated crimes, and crimes that “otherwise involves conduct that presents a serious potential risk of physical injury to another.” “Otherwise” implies some relationship between the residual clause and the enumerated crimes. But no one could figure out what it was. Should the Court determine the least risky of the enumerated crimes—either burglary or extortion, no one knew which—and determine whether the unenumerated crime was as risky as that? Should it identify some common characteristic of the enumerated crimes and check whether the unenumerated crime possesses that characteristic? If so, what characteristic?
Courts were not acting as courts. This, to me, is what Johnson boils down to. The Supreme Court cast Johnson as a due process decision: the Court held that “the indeterminacy of the wide-ranging inquiry required by the residual clause both denies fair notice to defendants and invites arbitrary enforcement by judges.” This seems like a curious justification in the context of criminal sentencing. Historically, criminals never had “fair notice” of their sentences. Criminal statutes would impose sentencing ranges of “a year to life,” and judges would have almost complete discretion to impose any sentence within that range. And sentences were often totally arbitrary—a judge might read a news story about violent crime and give 20 extra years to deter future offenders, and the sentence would be upheld on appeal. Modern sentencing reforms have led a more predictable sentencing process, but no one doubted that the old method of sentencing was constitutionally permissible. Justice Scalia—Johnson’s author— certainly did not.
So if “a year to life” schemes were constitutional, why wasn’t ACCA’s residual clause constitutional? In my view, it is because judges applying ACCA simply weren’t judging. Historically, judges did many things. They interpreted legal texts. They discerned, and extended, common law. They found historical facts. They applied law to fact. They managed criminal trials. They exercised discretion to impose just criminal sentences. But applying ACCA to unenumerated crimes didn’t fall into any of these categories. In theory, applying ACCA presented a question of law. In practice, there was nothing law-like about applying ACCA. Deciding whether possession of a sawed-off shotgun presented a “serious potential risk,” that was somehow comparable to the risk presented by burglary and/or extortion, cannot plausibly be described as the interpretation of a legal text or any other traditional judicial task.
At core, Justice Scalia was a traditionalist. To him, “due process of law” required some kind of traditionally judicial process. Imposing criminal sentences based on a process completely alien to judging as practiced for 200 years was not due process of law.
It’s an idea. It’s an invention. It’s an idea and an invention.
(This is Dall-E 2’s effort to merge burglars and software in one image. Not bad actually.)
Section 101, as applied to software patents, is as vague as ACCA. It’s not just ambiguous in close cases. It is flat-out incapable of being applied. No one knows what “abstract” means, no one knows what test to use, and cases boil down to random gestalt judgments.
No one knows how to determine whether software is an “abstract idea” or an “invention.” Let’s return to Visual Memory. The majority thought the claim was directed to concrete technology—improved computer memory. The dissent thought the claim was directed to an idea—categorical data storage. Both statements are true. The claim was both technology and an idea.
Is the Visual Memory patent an abstract idea because it contains a claim limitation that tells a computer to do something? Software always involves telling computers to do things. That’s what software is: instructions to a computer. Does this suggest that all software patents are ineligible? Or just software patents where the relevant software is described at a particularly high level of generality? If so, how high is too high? What test is used to determine height?
Suppose you think the claim in Visual Memory is an unpatentable abstract idea because it’s directed to a generic programmable memory. Let’s look at another claim in the same patent. The specification discusses using memory in “fast page mode,” in which a row in a memory page is accessed without having to continually respecify the row address, thereby reducing access time. Claim 6 covers the fast page mode embodiment. Here’s what it says:
6. A computer memory system connectable to a processor and having one or more programmable operational characteristics, said characteristics being defined through configuration by said computer based on the type of said processor comprising:
a main memory, connected to said processor by a bus, including a plurality of pages of a first type and a plurality of pages of a second type; and
a register connected to said memory for holding a page address of a most recently accessed one of said first type pages;
wherein said page address is used to reopen the most recently accessed one of said first type pages after one of said second type pages has been accessed; and
wherein a programmable operational characteristic of said system determines a type of data stored in said main memory.
This seems a bit more specific, but it still includes that “programmable operational characteristic” language. And it talks about “using” an address to “reopen” a page, which kind of sounds like something a human could do in a notebook. Idea or invention?
It’s not just Visual Memory. Take Finjan, Inc. v. Blue Coat Systems, Inc., 879 F.3d 1299 (Fed. Cir. 2018). Here’s what the claim at issue says:
1. A method comprising:
receiving by an inspector a Downloadable;
generating by the inspector a first Downloadable security profile that identifies suspicious code in the received Downloadable; and
linking by the inspector the first Downloadable security profile to the Downloadable before a web server makes the Downloadable available to web clients.
A “downloadable” was defined in the patent as a piece of software that’s downloaded. The claim is an idea: download software, scan it, and attach the results of that scan before sending it elsewhere. The claim is an invention: a new way of providing computer security. It’s an idea and an invention.
The Supreme Court has recently called for the Solicitor General to file briefs in two cases addressing Section 101. In one of them, Interactive Wearables LLC v. Polar Electric Oy, here’s what the independent claim says:
A content player comprising:
a receiver configured to receive content and together with the content information associated with the content,
a processor coupled to the receiver and configured to process the content and the information associated with the content,
memory coupled to the processor,
a first display coupled to the processor, and
playing device equipment coupled to the processor and configured to provide the content to a user of the content player, the playing device equipment comprising an audio player;
wherein the content player is a wearable content player configured to be controlled by a wireless remote control device comprising a second display, the wireless remote control device being configured to receive commands directing operations of the wearable content player, and wherein the wireless remote control device is configured to provide to the user at least a portion of the information associated with the content.
This isn’t exactly software architecture, but there’s software in there, and it presents a similar problem. You’ve got a content player and a wearable remote control with a monitor that displays information about the content. Idea or invention? It’s the idea of hooking up a remote control to a content player. It’s the invention of a new content player that allows you to see information about the content in real time.
I conclude there is no manageable way of distinguishing between software ideas and software inventions.
No one knows what the test is. Taking a step back, part of the problem is that the Federal Circuit can’t settle on a test for distinguishing abstract ideas from concrete inventions.
For example, the Federal Circuit frequently hears appeals in cases where the district court invalidates a patent under Section 101 at the motion to dismiss stages. In these cases the Federal Circuit sometimes decides patent eligibility as a matter of law, and sometimes remands for claim construction and fact-finding. No one seems to know what is supposed to happen on remand in these cases. I have yet to see a single concrete example of how patent eligibility can turn on claim construction. It is hard to even come up with a hypothetical case in which a claim would be an abstract idea if an ambiguous claim term was construed one way, and an invention if the claim term was construed another way.
Same with fact-finding. What facts should the court find? Alice includes this language:
“Using a computer to create and maintain ‘shadow’ accounts amounts to electronic recordkeeping—one of the most basic functions of a computer. … The same is true with respect to the use of a computer to obtain data, adjust account balances, and issue automated instructions; all of these computer functions are well-understood, routine, conventional activities previously known to the industry. In short, each step does no more than require a generic computer to perform generic computer functions.”
(Cleaned up.). The Court did not suggest fact-finding was needed on these conclusions, which is sensible because they are obviously correct. Perhaps in some cases, fact-finding is needed on whether “computer functions are well-understood, routine, conventional activities previously known to the industry.” But is this standard obviousness and anticipation fact-finding, or something else? Also, if the claim covers a genuine, non-anticipated, non-obvious advancement in computer technology, why are we even talking about abstract ideas? Nine years after Alice, no one has the slightest idea what fact-finding is supposed to happen or how those factual findings intersect with the ultimate legal eligibility analysis.
And, what’s the relationship between abstractness and functional claiming—that is, claims directed to a function, rather than an implementation of that function, the issue currently before the Supreme Court in Amgen? The Federal Circuit sometimes observes that “a claim that merely describes an effect or result dissociated from any method by which it is accomplished is not directed to patent-eligible subject matter.” Apple, Inc. v. Ameranth, Inc., 842 F.3d 1229, 1244 (Fed. Cir. 2016); see also Two-Way Media Ltd. v. Comcast Cable Commc'ns, LLC, 874 F.3d 1329, 1337 (Fed. Cir. 2017) (ineligible claim is written in “result-based functional language” that “does not sufficiently describe how to achieve these results in a non-abstract way”). Is functional claiming a requirement for an “abstract idea”? A factor that should be considered in deciding whether something is an “abstract idea”? A feature that correlates with something being an “abstract idea”? One can easily imagine abstractness without functional claiming (e.g., the mathematical algorithm in Bilski), or functional claiming without abstractness (“a car that performs the function of flying”), so it is not clear how the two concepts fit together.
As with ACCA’s residual clause, I don’t think judges are engaging in judging when they decide whether software patents are subject-matter eligible. Judges stare at the patent claim and specification, form an intuitive assessment of whether the patent is abstract or not, and decide whether to invalidate it based on that intuitive assessment. There’s no test, no law, no evidence, just rough justice. I don’t think it’s sustainable.
Enablement is the future
I don’t have any insight into what the Justice Department will recommend to the Supreme Court in the pending Section 101 cases, or how the Supreme Court would rule if it took such a case. But I suspect that the Justice Department, and especially the Supreme Court, may lack patience with the Federal Circuit’s jurisprudence in this area. The Court may hold that, other than patents that are directed to purely intellectual and non-technological concepts such as hedging, any patents involving software are patent-eligible.
I don’t relish this future. Many, perhaps most, software patents are terrible, serving as a drain on innovation without contributing anything to the public store of knowledge. But those who oppose such patents need a legal strategy beyond reflexive use of Section 101.
Amgen—if it comes out the right way—provides some hope. Enablement, at least under Sanofi’s position, is laser focused on functional claiming. For a functional claim to be enabled, all ways of accomplishing that function must be disclosed in the specification. Courts should isolate the invention’s point of novelty—the claimed improvement over the prior art—and assess whether the invention claims a result, or whether instead it claims a way of obtaining it. If it claims the result, and the specification describes one way (or zero ways) of accomplishing that result, then the claim is not enabled, and can be invalidated without considering subject-matter eligibility.
Patent lawyers often object that enablement is a less useful defense than ineligibility because enablement typically requires full discovery, whereas ineligibility can be decided at the motion to dismiss stage. Well, we can change that rule. There is no edict from God requiring full discovery in every enablement case. If it is clear at the motion to dismiss stage that the patent isn’t enabled, it can be invalidated. In Visual Memory, the Federal Circuit could easily have decided prior to discovery that the specification did not enable every way of programming a memory to align with a processor’s operation.
Enablement doesn’t solve every problem. Many cases will still be difficult, because of challenges in identifying a “function,” determining the point of novelty in a claim, interpreting the specification, and others. But at least the inquiry is tethered to a statutory standard and courts will have some sense of what they are trying to do: determine whether the inventor has actually invented everything she seeks to monopolize.
Comments are welcome! In next week’s post, I will discuss a forthcoming Supreme Court case, Samia v. United States, and the constitutional law of evidence.
As always, these posts reflect my own views, not the views of Jenner & Block.
More precisely, a non-quantum algorithm that factors large integers in polynomial time.