“Since the Section 112 Guidelines were promulgated, the PTAB has issued several decisions finding often-used computer-implemented terms to be indefinite for lacking an algorithm; terms which have been widely accepted and commonly-used by software applicants to delineate the very existence of an algorithm.”
When the USPTO issued its 2019 Revised Patent Subject Matter Eligibility Guidance in January of this year, it seemed as if the patentability tides had finally shifted in favor of software applicants. Far less attention and fanfare, however, was afforded to the concurrently issued and unassuming Section 112 Guidelines on examination practice for computer-related and software claims. In particular, potential pitfalls awaiting software applicants may lie unforeseen in the requirement that “[f]or a computer-implemented 112(f) claim, the specification must disclose an algorithm for performing the claimed computer function, or else the claim is indefinite.”
A Closer Look
For claims directed to computer-implemented steps, generic terms (so-called “nonce terms” such as “unit” or “module”) coupled with functional claiming is almost unavoidable. So, the examination of 112(f) claims under the Section 112 Guidelines deserves a closer look. Notably, since the time that the Section 112 Guidelines were promulgated, the Patent Trial and Appeal Board (PTAB) has issued several decisions finding often-used computer-implemented terms (“determin[ing]”, “process[ing]”, “analyz[ing]”, “evaluat[ing]”) to be indefinite for lacking an algorithm; , terms which have been widely accepted and commonly-used by software applicants to delineate the very existence of an algorithm. In several recent decisions at the PTAB, terms such as “determining” and “evaluating” are deemed purely functional terms that require an algorithm for support.
Although the requirement for an algorithm appears unavoidable under the Guidelines, a point in favor of the applicant is the clear delineation that an algorithm is not limited to disclosing a flow chart. Rather, an algorithm is defined as “a finite sequence of steps for solving a logical or mathematical problem or performing a task.” An applicant may “express that algorithm in any understandable terms including mathematical formula, in prose, flow chart, or any other manner that provides sufficient structure.”
However, the difficulty for many appellants facing 112(f) claim interpretations at the PTAB is that, under the new Section 112 Guidelines, an algorithm must be sufficient to perform the entire claimed function(s).
Ex Parte Callier
For example, in Ex Parte Callier, No. 2018-000321 (P.T.A.B. June 19, 2019), Appellant claimed:
A system, comprising:
a test device comprising:
(a) a label zone comprising
(i) a first population of mobilizable, detectable particles for specific binding to a test analyte in a sample; and
(ii) a second population of mobilizable, detectable particles for specific binding to a non-test analyte but not to the test analyte; and
(b) a first detection site at a first position comprising a first binding member having specific binding affinity for detectable particles in the first population bound to the test analyte; and
(c) a second detection site at a second position that is separate from the first position, the second detection site comprising a second binding member having specific binding affinity for detectable particles in the second population bound to the non-test analyte; an analyzer comprising:
(1) a drawer dimensioned to receive the test device,
(2) an optical system for detection of a signal generated from the first population of particles at the first position and the second population of particles at the second position when each population reaches a specific and separate position on the test device,
(3) a processor and memory resources for control of a motor and for processing data from an optics module, wherein the processor and memory resources are configured to evaluate a sample using a ratio of signal detected from all or a portion of the second population of particles to signal detected from all or a portion of the first population of particles (i) to determine a time insensitive cut-off value for the analyzer, and (ii) to determine the presence or absence of analyte in the sample.
The Examiner found that Appellants’ Specification “discloses no corresponding algorithm associated with ‘a processor and memory resources [which are configured to evaluate a sample]”‘. The Appellants countered that an algorithm was shown because the specification disclosed that the processor and memory are configured with an algorithm stored in the memory. But that argument did not work, and the claim was ultimately found indefinite for lack of a “sufficient algorithm associated with ‘a processor and memory resources’”.
Here’s my question; why didn’t the Appellants just say that the algorithm is in the claim? “[C]ontrol of a motor and for processing data from an optics module” was immediately followed by “wherein”. Thus, shouldn’t all the limitations following “wherein” (i.e., evaluate . . . , determine . . . , determine . . .) be reasonably interpreted to modify what came before “wherein”? I would think so. In Callier, the claimed function of control[ing] of a motor and processing data is accomplished via the algorithm recited, which is evaluating a sample, determining a cut-off value, and determining the presence of an analyte. There should have been no requirement for the “evaluate” step and the “determine” steps to be supported by an algorithm whatsoever, as those were not functions.
Ex Parte Liu
A similar issue arose in Ex Parte Liu, No. 2018-000996 (P.T.A.B. June 4, 2019), in which Appellant claimed:
An apparatus for detecting target landmarks in a large volume, the method comprising:
means for detecting a set of landmark candidates for each of a plurality of target landmarks in an image volume;
means for generating a node potential value for each landmark candidate based at least in part on an error value determined using spatial histogram-based error regression;
means for generating edge potential values for pairs of landmark candidates based on a bifurcation analysis of the image volume using vessel tracing; and
means for determining an optimal landmark candidate for each target landmark, using a Markov random field model based on the node potential values and the edge potential values.
The PTAB found that the specification merely states that the Markov Random Field model analyzes landmark candidates to determine optimal landmark candidates based on the node potential values and edge potential values, but failed to disclose any algorithm for the analysis or how the optimal landmark candidate is determined. In other words, the support in the specification was basically a repetition of what was already in the claim, thus, no algorithm was found and the claim was indefinite.
But what if the claim merely recited, “means for determining an optimal landmark candidate for each target landmark” and kept it at that? What if the claim made no mention of “using a Markov random field model based on the node potential values and the edge potential values”? Then once the USPTO demanded an algorithm for determining an optimal landmark candidate, the Appellant could have relied on the specification’s teaching of using Markov Random Field Model. That’s not repeating the claim!
The sad thing about these cases is that the applicants are being punished for writing thorough claims—for reciting more features that are necessary and finding themselves in the position of having to provide further support for claims that are so definite they are wrong. It is hard to tell if this trend preceded the new Section 112 Guidelines or if we are seeing the effects of the new Section 112 Guidelines illustrated in these cases. Whatever the case, the Section 112 Guidelines are pretty explicit on the need for an algorithm(s) to support the entirety of computer-implemented claims. Therefore, one could venture to assume that this requirement could only make claiming even more of a word game for software applicants.
Image Source: Deposit Photos
Image ID: 2597542
Join the Discussion
6 comments so far.
AnonAugust 23, 2019 04:40 pm
Sister Anon (not the usual Anon),
Your hyperlink at post 4 redirects to a the USPTO.gov website.
Was this intentional?
PatrickAugust 22, 2019 08:43 pm
I have not looked at the prosecution history, but one reason applicants may not have pointed to the claims to provide support for the algorithm is that the claims on appeal may not have been the originally filed claims. If the claims on appeal were the originally filed claims, I could see an argument that the claims did in fact provide the algorithm. This of course would still leave a problem of lack of antecedent basis in the specification for the claim language…but that should be easily remedied with an amendment to the specification to include the claim language.
AnonAugust 22, 2019 02:16 pm
TFCFMAugust 22, 2019 09:55 am
My two cents RE what’s presented about Liu:
The post asserts that:
“The PTAB found that the specification merely states that the Markov Random Field model analyzes landmark candidates to determine optimal landmark candidates based on …, but failed to disclose any algorithm for the analysis or how the optimal landmark candidate is determined…”
“But what if the claim merely recited, “means for determining an optimal landmark candidate for each target landmark” and kept it at that? … Then once the USPTO demanded an algorithm for determining an optimal landmark candidate, the Appellant could have relied on the specification’s teaching of using Markov Random Field Model.”
It seems to me that the Examiner asserts that the claim does not adequately describe the metes and bounds of the apparatus that is claimed (in that the algorithm embodied in the apparatus is indefinite in scope).
Whether the MRF model is written expressly in the claim or read into it when construing the “means for determining” term seems to me to make no difference. The relevant question, definiteness-wise, is whether a skilled artisan can tell the metes and bounds of the “metes for determining” merely from recitation of the MRF model.
I have no relevant knowledge of the MRF model, but if it’s like most mathematical models, there are many (infinite number?) ways of implementing it. The relevant question, then seems to reduce to whether a skilled artisan would be able to clearly and distinctly determine that which algorithms are within the scope of the claim and which are beyond it. Simply naming a model seems to me unlikely to provide such clarity. If the applicant were able to demonstrate that a skilled artisan would know *exactly* which algorithms are “in” and which are “out,” my opinion would be different.
Still, at the end of the day, it seems to me immaterial whether or not the algorithm is merely named in the claim and extremely material whether the name of the algorithm (expressed or imported) provides a clear, definite boundary to the claim.
Remember, definiteness is all about “my patent starts *here* and stops *there*.”
AnonAugust 22, 2019 07:03 am
The clear choice of using “means for” was very much trending as you indicated.
The courts (and here, reflecting that trend with the pseudo-court of the administrative agency) are not content with what “scriveners” are doing, so they are rewriting the claims as they want them to be.
CuriousAugust 21, 2019 07:15 pm
I reviewed both decisions and Ex parte Liu stood out to me because applicants used the “means for” language, which the Board entered a new grounds of rejection to reject these claims under 35 USC 112.
I thought any self-respecting patent attorney avoided “means for” language. Means for language was discouraged before the turn of the millennium.