Semiconductor Patents

Via Licensing was one of the original patent pool administrators for wireless standards. Merged with Avanci in 2022. Avanci now administers pools for 2G/3G/4G/5G cellular patents, licensing to IoT device manufacturers and automakers (Vehicle 5G Connectivity) at fixed per-unit rates. Avanci's automotive licensing model (fixed royalty per connected vehicle regardless of number of cellular SEPs used) has been controversial but adopted by most major automakers.

JEDEC (Joint Electron Device Engineering Council) requires member companies to disclose patents they know to be essential to JEDEC standards and to commit to licensing those patents on RAND (Reasonable and Non-Discriminatory) terms. JEDEC does not operate a formal patent pool — each member licenses its SEPs individually. The Rambus disputes (in which Rambus participated in JEDEC while allegedly concealing SEPs, then withdrew and asserted them) led to major FTC enforcement action and shaped JEDEC's current disclosure requirements.

The IEEE Standards Association (IEEE-SA) patent policy requires that for essential patents declared to IEEE, the SEP holder commit to one of three levels of licensing obligation: (LOA Option 1) reasonable rates, terms, and conditions (RAND); (LOA Option 2) not to enforce essential patent claims; or (LOA Option 3) reciprocal licensing. The 2015 IEEE-SA policy update to define 'reasonable rates' by excluding the value of the standard itself from the royalty calculation was controversial — Qualcomm and InterDigital objected; Ericsson and Nokia adopted the updated policy. Wi-Fi chipset manufacturers (Broadcom, Qualcomm Atheros, Intel, MediaTek) must navigate IEEE-SA SEP licensing from multiple holders.

ETSI (European Telecommunications Standards Institute) and 3GPP (3rd Generation Partnership Project) set the cellular standards implemented in all LTE and 5G chips. ETSI's IPR policy requires SEP holders to commit to FRAND licensing before the standard is published (ETSI IPR Policy, Clause 6.1). ETSI does not define FRAND rates — that is left to negotiation and court determination. The battle over FRAND rate-setting methodology for 5G SEPs has been litigated globally: UK (Unwired Planet v. Huawei [2020] UKSC 37 — UK courts can set global FRAND rates), EU (Huawei v. ZTE C-170/13 [2015] — framework for FRAND injunctions), US (FTC v. Qualcomm). FRAND licensing costs are a significant, non-optional line item in chipset gross margin.

A single semiconductor chip typically has multiple layers of IP protection from multiple different IP owners. The main categories are: (1) Process patents: cover the manufacturing steps used to fabricate the chip — lithography, doping, deposition, etch. Owned by foundries (TSMC, GlobalFoundries) and IDMs (Intel, Samsung). Cannot be designed around at a given node; (2) Mask work protection under the Semiconductor Chip Protection Act of 1984 (17 U.S.C. §§ 901–914): covers the 3D arrangement of circuit elements in the chip layout. 10-year term. Reverse engineering for study is permitted; (3) Architecture patents: cover the microarchitecture of the processor or chip design — pipeline, cache hierarchy, branch prediction, memory system. These are the most litigated chip patents; (4) Standard-essential patents (SEPs): patents essential to implementing wireless or memory standards (5G/LTE for modem chips; DDR for memory interfaces). Must be licensed on FRAND terms; (5) EDA and design methodology patents: cover tools used to design chips — owned by Synopsys, Cadence, Siemens EDA. EDA tool licenses don't immunize customers from third-party architecture patents. A fabless chip startup must navigate all five layers — particularly process patents (through foundry indemnification), architecture patents (FTO analysis), and SEPs (mandatory FRAND licensing if building standards-compliant wireless chips).

The CHIPS and Science Act (Pub. L. 117-167, 2022) provides $52.7 billion for domestic semiconductor manufacturing and R&D, but it comes with significant IP-relevant constraints. Manufacturing grants (the $39B fabrication incentive) come with 10-year 'guardrails' prohibiting recipients from expanding semiconductor manufacturing capacity in 'countries of concern' (primarily China) and from certain technology licensing with foreign entities of concern. This directly limits the IP licensing strategy of CHIPS grant recipients — they may be unable to license process technology to Chinese foundries or joint-venture partners for a decade. R&D grants (NSTC, NAPMP, university programs) come with Bayh-Dole-style provisions: recipients own the resulting patents but the US government receives a royalty-free license and retains march-in rights if the invention is not commercialized. CHIPS Act recipients should structure their R&D efforts carefully — work funded by CHIPS R&D grants will be subject to these government license and march-in provisions. The 25% Advanced Manufacturing Investment Tax Credit (§ 48D AMITC) is tax-side and does not directly affect patent ownership, but CHIPS funding guardrails and the tax credit eligibility rules together constrain where subsidized manufacturing capacity can be deployed and licensed.

FRAND stands for Fair, Reasonable, and Non-Discriminatory — the licensing commitment that holders of standard-essential patents (SEPs) must make when their patents are declared essential to a telecommunications or technology standard. For semiconductor companies, FRAND licensing matters most for wireless chipsets (5G/LTE modems, Wi-Fi chips, Bluetooth) and memory interfaces. Here's how it works: (1) A standards body (ETSI for cellular, IEEE-SA for Wi-Fi/Ethernet) develops a standard (5G NR, 802.11ax). Participating companies declare patents they believe are essential to implementing the standard; (2) In exchange for the standard adopting their technology, declared SEP holders commit to license those SEPs on FRAND terms to any implementer who requests a license; (3) An implementer who needs to build a 5G modem chip CANNOT design around SEPs — by definition, implementing the standard requires practicing the SEPs. They must negotiate a FRAND license; (4) If FRAND rates can't be agreed upon, courts can determine them (UK Unwired Planet v. Huawei; US courts; EU member state courts). Courts have increasingly been willing to set global FRAND royalty rates. For a fabless company building 5G chipsets, FRAND licensing is a mandatory cost of goods — not optional. Major 5G SEP holders (Qualcomm, Nokia, Ericsson, Huawei, Samsung, InterDigital) actively license chipset manufacturers, often at per-unit royalty rates. The chipset manufacturer must model FRAND licensing costs as a non-negotiable line item in chip gross margin before committing to a wireless standard-compliant product.

ARM Holdings (now owned by SoftBank and publicly traded as ARM) licenses its intellectual property — primarily processor architecture and physical IP — to almost every major chip company in the world. There are three main ARM license types: (1) Architecture License (ALA): authorizes the licensee to design custom processor cores that implement the ARM instruction set architecture (ISA). Holders of ALAs include Apple (M-series, A-series), Qualcomm (Oryon/Snapdragon X Elite), and Amazon (Graviton). ALAs give maximum design freedom — the licensee can create entirely custom microarchitectures that are ISA-compatible. ALAs have historically been fixed-fee with per-unit royalties, but ARM is transitioning ALA pricing to include royalties based on the chip's selling price or ACV, creating significant cost uncertainty for ALA holders; (2) Processor License (PLA): authorizes the licensee to implement a specific ARM-designed processor core (e.g., Cortex-A78, Cortex-X4) in a product. The licensee does not design the core — they take ARM's RTL (register-transfer level) implementation and integrate it. PLA royalties are per-chip; (3) Physical IP License: access to ARM's optimized standard cell libraries, IO pads, and other physical components — tied to specific foundry process nodes. Risks for chip companies: (a) License transferability — ARM licenses are generally non-transferable; acquiring a company with an ARM license (e.g., Qualcomm's acquisition of Nuvia) may not automatically transfer the license, as the Arm v. Qualcomm litigation (D. Del., filed 2022) illustrates; (b) Royalty escalation — ARM is actively pushing for royalty increases that could significantly raise chip COGS for existing licensees; (c) RISC-V as an alternative — the open ISA RISC-V eliminates ARM ISA-level licensing risk (no patent license needed for the ISA) but implementations may still face architecture patent risk on specific microarchitectural choices.

Yes — and for most well-funded fabless startups, building a patent portfolio is strategically important even if the startup has no current plans to litigate. Here's why: (1) Defensive cross-licensing: the semiconductor industry is a dense patent thicket. When a large chip company alleges infringement (or sends a licensing demand), having a portfolio of your own patents gives you chips to negotiate with. Most major chip company disputes are resolved through cross-licenses — 'you license my patents, I license yours, we both walk away.' Without your own portfolio, you have nothing to trade; (2) Fundraising and acquisition signaling: patent portfolios are tangible IP assets that appear on the balance sheet (if capitalized) or are disclosed in due diligence. VCs and acquirers use patent counts as a proxy (imperfect, but real) for technical differentiation and defensibility. Chip startups have been acquired primarily for their patent portfolios even when the product didn't succeed; (3) ITC § 337 exclusion orders: if a competitor imports infringing chips into the US, a patent portfolio enables an ITC Section 337 complaint — which can result in an import exclusion order faster than district court injunctions; (4) Freedom to operate: the process of building a portfolio (prior art searches, claim drafting) naturally surfaces third-party patents that the startup's chip might infringe — identifying FTO risks before tapeout is dramatically cheaper than post-tapeout design changes. For the portfolio strategy: focus first on the architecture and algorithm inventions that are core to the chip's differentiated performance — these are the claims most likely to have licensing value. Process patents are not available to fabless companies. SEPs require participating in standards bodies. The most valuable fabless portfolio consists of architecture patents covering what makes your chip faster, more efficient, or more capable than competitors.