1:24-cv-00989
Kamal v. Femtosense Inc
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: Andrew Magdy Kamal (Michigan)
- Defendant: Femtosense, Inc. (Delaware) and Sam Fok (California)
- Plaintiff's Counsel: Law Office of Andrew Magdy Kamal
- Case Identification: 1:24-cv-00989, D. Del., 09/30/2025
- Venue Allegations: Venue is based on Defendant Femtosense, Inc. being a for-profit corporation located in California but incorporated in Delaware.
- Core Dispute: Plaintiff alleges that Defendants' artificial intelligence technology, which utilizes data compression, infringes a patent related to a data compression method.
- Technical Context: The technology at issue involves methods for compressing large datasets, a critical function for enabling efficient artificial intelligence and machine learning applications, particularly on resource-constrained hardware.
- Key Procedural History: The complaint states the matter was removed from the Delaware Chancery Court. Plaintiff alleges having placed Defendants on notice of infringement via a cease and desist letter prior to filing suit. The complaint also includes non-patent causes of action for willful patent infringement, defamation, duress, and violations of civil rights.
Case Timeline
| Date | Event |
|---|---|
| 2018-08-09 | U.S. Patent 10,965,315 Priority Date |
| 2021-03-30 | U.S. Patent 10,965,315 Issued |
| 2024-04-15 | Plaintiff allegedly sent Defendants a cease and desist letter |
| 2024-05-25 | Plaintiff filed Proof of Service in a related California action |
| 2024-06-26 | Defendants allegedly filed their first answer in the California action |
| 2024-07-17 | Date of an email from Defendants' counsel alleged to be threatening |
| 2025-09-30 | Amended Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 10,965,315 - Data Compression Method
Issued March 30, 2021 ('315 Patent)
The Invention Explained
- Problem Addressed: The patent's background describes the challenge of processing and storing vast amounts of data generated by modern systems, noting that conventional data compression methods can be inefficient '315 Patent, col. 1:20-33 The goal is to provide a more effective method and system for compressing data '315 Patent, col. 1:17-19
- The Patented Solution: The invention is a method for compressing a data set by categorizing its values into a first category ("complexities") or a second category ("memoryless data") Compl. ¶9 Values in the first category are added to the compressed data set, while values in the second are excluded '315 Patent, abstract However, these excluded values are used to update a separate statistical distribution of memoryless data '315 Patent, abstract '315 Patent, Fig. 1 The process operates in two phases: a first phase where categorization is based on predefined criteria, and a subsequent second phase where categorization is based on the updated statistical distribution, potentially using Bayesian techniques '315 Patent, abstract Compl. ¶¶11-12
- Technical Importance: The method is described as enhancing data compression efficiency and accuracy, which is crucial for data-intensive applications like real-time analytics Compl. ¶8 '315 Patent, col. 1:28-33
Key Claims at a Glance
The complaint asserts infringement of independent claims 1 and 25 Compl. ¶17
- Independent Claim 1 (Method):
- obtaining a data set and criteria for determining whether individual values correspond to a first or second category;
- determining that some values correspond to the first category and others to the second;
- based on a value corresponding to the first category, adding the value to a compressed data set;
- based on a value corresponding to the second category, excluding it from the compressed data set and updating a statistical distribution of values of the second category;
- wherein the determination occurs in a first phase based on comparison to criteria, and a subsequent second phase based on the statistical distribution.
- Independent Claim 25 (Computing Device):
- A computing device with memory and a processing circuit configured to perform the steps of claim 1.
The complaint alleges infringement of dependent claims 2-17 and 20-24, which add further limitations such as using quadtree data structures and Riemann zeta function verification Compl. ¶31 Compl. ¶21
III. The Accused Instrumentality
Product Identification
The accused instrumentalities are Defendants' "products and/or methods" that embody the patented invention, specifically Femtosense's technology for "efficient, scalable, and affordable AI" Compl. ¶14 Compl. ¶16 The infringement analysis in the complaint's exhibit points to specific products like the "AI-ADAM-100 MCU" Doc. 38-2, p. 17
Functionality and Market Context
The complaint alleges that Femtosense's technology uses "sparse, localized mathematics" for data compression to enable its AI solutions Compl. ¶16 The accused method is alleged to involve categorizing data into a first category ("Tuple") and a second ("IOTARGET"), performing a "quantization operation" on the first category, and utilizing a two-phase process Compl. ¶¶18-20 An image provided in the complaint's exhibit depicts Femtosense's AI-ADAM100 chip, which is described as a "SPU + MCU for on-device voice cleanup and flexible commands" Doc. 38-2, p. 17 The complaint's exhibit also includes a screenshot of Femtosense's marketing material describing its technology's use of "Sparsity-Aware-Training (SAT)" which "uses compression techniques to simplify the model" Doc. 38-2, p. 7
IV. Analysis of Infringement Allegations
The complaint references Exhibit B, which contains a claim chart analysis that was not fully reproduced in the provided documents but whose core allegations are summarized below based on the exhibit's contents.
'315 Patent Infringement Allegations (Claim 1)
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| obtaining a data set and criteria for determining whether individual values from the data set correspond to a first category or a second category of values; determining that some values... correspond to the first category, and that other values... correspond to the second category; | Femtosense's method is alleged to obtain a data set and categorize it into a "first category of values: Tuple" and a "second category of values: IOTARGET". A code snippet from "spu_runner.py" is provided as evidence. | ¶18 | col. 11:53-61 |
| based on one of the values corresponding to the first category, adding the value to a compressed data set; | The accused method is alleged to perform a "quantization operation on a first category of data" to add values to a compressed set. | ¶19 | col. 11:62-64 |
| based on one of the values corresponding to the second category; excluding the value from the compressed data set; and updating a statistical distribution of values of the second category in the data set based on the value; | The accused method is alleged to exclude a second category of data. A code snippet showing an "assert False" for an "unknown message type" is presented as evidence of this exclusion. | ¶19 | col. 12:1-5 |
| wherein during a first phase, the determining is performed for a plurality of values from a first portion of the data set based on comparison of the values to the criteria; | This is alleged to be met by a "hw_send" function in Femtosense's code, interpreted as "sending data to the hardware for first portion of data." | ¶20 | col. 12:5-10 |
| and wherein during a second phase that is subsequent to the first phase, the determining is performed for a plurality of values from a second portion of the data set that is different from the first portion based on the statistical distribution. | This is alleged to be met by a function for "Dequantize outputs," which is interpreted as performing the second phase determination. | ¶20 | col. 12:11-16 |
'315 Patent Infringement Allegations (Claim 25)
The allegations for Claim 25 mirror those for Claim 1, but are directed at a "computing device" configured to perform the method steps Compl. ¶17 The complaint's exhibit presents an image of the "AI-ADAM100" chip as the infringing device, which is described as a "Multi-Chip Package (M(2))" with a "SPU-001 NPU and Cortex-M0+ MCU" Doc. 38-2, p. 17 The allegations assert this device is configured to perform the categorization, compression, and two-phase operations described above Doc. 38-2, pp. 20-26
- Identified Points of Contention:
- Scope Questions: The complaint alleges that Femtosense's "quantization" Compl. ¶19 and "dequantization" Compl. ¶20 operations correspond to the patent's specific two-phase categorization and compression method. A central question will be whether these general industry terms, as used by the accused product, practice the specific sequence and logic required by the claims. For example, does "sending data to external hardware" Compl. ¶20 perform the claimed function of the "first phase"?
- Technical Questions: The complaint alleges that Femtosense's code categorizes data into "Tuple" and "IOTARGET" Compl. ¶18, and that this maps to the patent's "first category" and "second category". A key technical question is whether the accused system's handling of "IOTARGET" data matches the claim requirement of being excluded from the compressed set but used to "updat[e] a statistical distribution." The evidence provided in the complaint focuses on exclusion but does not explicitly show how a statistical distribution is updated or used in a second phase as claimed.
V. Key Claim Terms for Construction
The Term: "statistical distribution"
Context and Importance: This term is central to the claimed two-phase process. The second phase determination is performed "based on the statistical distribution" that was updated by values from the "second category" '315 Patent, abstract The infringement theory hinges on whether the accused product uses such a distribution in the claimed manner. Practitioners may focus on this term because the complaint's allegation for the second phase points to a "dequantization" step Compl. ¶20, and it is not facially apparent how this relates to a determination based on a previously updated statistical distribution.
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The patent does not appear to define "statistical distribution" in a special way, suggesting it could be given its plain and ordinary meaning in the context of data science. The specification mentions that the update can simply reflect "a quantity of times the value has been found" '315 Patent, col. 13:12-13
- Evidence for a Narrower Interpretation: Figure 1 explicitly depicts the "Statistical Distribution of Memoryless Data" as a distinct output of the process, separate from the "Compressed Data Set" '315 Patent, Fig. 1 The detailed description ties its use to the second phase determination via Bayes' Theorem '315 Patent, col. 8:3-17, which may support an interpretation requiring a structure capable of supporting such a probabilistic analysis.
The Term: "first category" and "second category"
Context and Importance: The patent defines a specific functional relationship between these categories: the first is compressed and stored, while the second is excluded but used to update a statistical distribution '315 Patent, abstract The complaint alleges that Femtosense's "Tuple" and "IOTARGET" map to these categories Compl. ¶18 The case may depend on whether "IOTARGET" data is handled in the specific manner required for the "second category."
Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The claims themselves do not name the categories, leaving them as functional placeholders. One could argue any data segregation where one part is stored and another is used for statistical purposes falls within the scope.
- Evidence for a Narrower Interpretation: The specification repeatedly refers to the categories as "complexities" and "memoryless data" '315 Patent, col. 5:12-16 The abstract itself introduces this terminology. A defendant may argue that these labels imply technical characteristics that the accused "Tuple" and "IOTARGET" categories do not possess.
VI. Other Allegations
- Indirect Infringement: The complaint makes conclusory allegations of induced and contributory infringement Compl. ¶¶32-33 but does not provide specific facts to support the requisite elements of knowledge and intent, such as identifying specific instructions or components for inducement or contributory infringement.
- Willful Infringement: The complaint alleges willful infringement based on Defendants' continued infringing activities after receiving notice of infringement via a certified letter dated April 15, 2024 Compl. ¶23 Compl. ¶36 This allegation of pre-suit knowledge, if substantiated, may support a claim for willfulness.
VII. Analyst's Conclusion: Key Questions for the Case
- A core issue will be one of technical mapping: does the accused product's use of general concepts like "quantization," "dequantization," and code objects named "Tuple" and "IOTARGET" actually implement the specific, multi-step logic of the '315 patent's claims? The court will need to look beyond labels and determine if the accused method functionally aligns with the claimed two-phase process centered on a "statistical distribution."
- A second key question will be one of evidentiary proof for complex limitations: The complaint alleges that the accused technology incorporates "Bayesian Learning" and "Riemann Zeta Verification" Compl. ¶21 Given the mathematical sophistication of these techniques, a central challenge for the plaintiff will be to provide concrete evidence that the accused commercial product actually performs these specific, complex functions as required by the dependent claims, rather than using more generic data processing methods.