An embedded product — a connected Electronic Control Unit (ECU), a medical device controller, an industrial IoT sensor node, an autonomous vehicle compute platform — is simultaneously a hardware artifact, a firmware execution environment, a software application, and in many cases a trained machine learning model. Each of these layers has a distinct IP protection framework with different eligibility requirements, different scope of protection, different duration, and different enforcement mechanisms.
Most technology companies protect one layer well and neglect the others. Companies founded by hardware engineers tend to focus on patents for circuit innovations and overlook firmware trade secrets. Companies with software roots tend to protect algorithms and overlook the hardware implementation that makes the product competitive. A coordinated IP strategy covers all layers intentionally and maps each protection mechanism to the specific competitive advantages it is designed to preserve.
Layer 1: Hardware — Patents and Mask Work Rights
Novel hardware innovations — new circuit topologies, new sensor architectures, new packaging approaches, new PCB layout techniques that provide electrical performance advantages — are the natural domain of utility patents. A hardware patent that claims a specific circuit configuration can be extremely valuable because it is difficult to design around without sacrificing the performance advantage the circuit provides.
Hardware patent prosecution for embedded systems requires claim drafting that accurately captures the inventive concept at the right level of abstraction. Claims that are too narrow — tied to a specific component value or a specific technology node — are easy to design around. Claims that are too broad — capturing any circuit that achieves a functional result — face rejection for claiming an abstract idea or a natural phenomenon. The sweet spot requires a deep understanding of both the engineering and the claim drafting craft.
For companies that design custom silicon — Application-Specific Integrated Circuit (ASIC) or custom IP blocks on a System-on-Chip (SoC) — mask work rights under the Semiconductor Chip Protection Act of 1984 (SCPA or 17 U.S.C. § 901 et seq.) provides an additional layer of protection for the physical layout of the chip. Mask work protection prevents copying of the chip layout directly, though it does not prevent a competitor from independently designing a chip that performs the same function using a different layout. Registration with the United States Copyright Office is required within two years of first commercial exploitation.
Trade secrets are also applicable at the hardware layer for innovations that are not reverse-engineerable from the shipped product — novel manufacturing processes, specialized materials combinations, proprietary calibration procedures, and test and measurement methodologies that contribute to product performance.
Layer 2: Firmware — Copyright, Trade Secrets, and Patents
Firmware — the software that runs directly on the embedded processor and interfaces with the hardware — is protected by copyright as a literary work from the moment it is created. Copyright in firmware protects the specific expression: the source code as written, the compiled binary as expressed. It does not protect the functional behavior, the algorithms, or the system architecture.
The more practically significant protection for firmware is trade secret law. Firmware in a shipped product is typically stored in flash memory that is either physically inaccessible to most reverse engineers or read-protected by the microcontroller unit (MCU) field-programmable gate array's (FPGA) security features. Well-implemented read-back protection — using the debug port lockout, code protection fuses, and secure boot with encrypted firmware images — can make firmware trade secret protection robust even in a distributed product.
Novel firmware algorithms — particularly those that produce specific, concrete improvements in system performance, reliability, or safety (including Automotive Safety Integrity Level D (ASIL-D) systems whose Hazard Analysis and Risk Assessment (HARA) and Threat Analysis and Risk Assessment (TARA) documents form part of the legal safety case) — may also be patentable, subject to the 35 U.S.C. § 101 eligibility analysis discussed in the companion article Trade Secrets vs. Patents for Algorithms: A Decision Framework for Embedded Systems Companies. Real-time control algorithms, novel interrupt handling schemes, power management techniques, and communication protocol optimizations are all candidates for patent protection if they represent genuine technical advances.
A specific area worth flagging for automotive embedded systems companies: the Automotive Open System Architecture (AUTOSAR) software architecture, which is widely used in automotive ECU firmware, involves complex IP licensing arrangements between the AUTOSAR consortium members. Companies contributing to or implementing AUTOSAR must understand the scope of the AUTOSAR IP licensing terms and how they interact with proprietary firmware development built on the AUTOSAR platform.
Layer 3: Application Software — Copyright and Trade Secrets
For embedded products that include a higher-level software application layer — a user interface application, a cloud connectivity stack, a machine learning inference engine — the protection framework shifts toward software copyright and trade secrets, with patents playing a supporting role for novel algorithmic innovations.
The key IP strategy decision at this layer is which elements of the application software to pursue as open source — subject to a Software Bill of Materials (SBOM) audit — and which to protect as proprietary. Open-sourcing non-competitive elements of the software stack — device drivers for commodity hardware, utility libraries, example applications — can build developer community and ecosystem while allowing the company to focus IP protection on the genuinely differentiated elements.
Layer 4: Trained ML Models — Emerging IP Framework
For embedded products that incorporate trained machine learning models — neural networks for computer vision, anomaly detection for predictive maintenance, natural language processing for voice interfaces — the IP framework is actively evolving and presents significant strategic choices.
The training data used to train the model may be protectable as a trade secret if it is proprietary and not publicly available. The model architecture may be patentable if it represents a novel technical approach. The trained model weights are the subject of ongoing legal debate — some courts have treated them as trade secrets, others have been skeptical. The output of the trained model is subject to the same human authorship analysis that applies to other AI-generated content.
For automotive embedded systems specifically, the training data for safety-critical perception systems — labeled sensor data from real-world driving scenarios — represents one of the most valuable and competitively significant IP assets in the industry. Protecting that data as a trade secret, with appropriate access controls and contractual restrictions on use, is a high-priority IP strategy element.
guibert.law Approach to Multi-Layer IP Strategy
The most effective IP strategy for an embedded product maps each layer of the product architecture to the protection mechanism best suited to its characteristics — and coordinates filing timing, claim scope, and trade secret program elements to create overlapping, complementary protection. A competitor who designs around the hardware patent still faces the firmware trade secret. A competitor who reverse-engineers the firmware still cannot copy the training data. Layered protection is resilient protection.
Attorney advertising. The information in this post is provided for general informational purposes and does not constitute legal advice. Prior results do not guarantee a similar outcome. © 2026 guibert.law