Lattice Semiconductor Corporation (LSCC)

The low-power FPGA as the robot's sensor-fusion glue layer — explicit robotics design-ins via the NVIDIA Halos and TI partnerships — and unlike the mega-caps, robotics is a real growth vector, not a rounding error.

Investment Research · Photoncap-style deep dive · v1 of "Lattice Semiconductor" · May 14, 2026

What Lattice physically does

Lattice makes small, low-power field-programmable gate arrays — FPGAs that, unlike Intel's or AMD's data-center FPGAs, are measured in milliwatts and millimeters rather than watts and centimeters. An FPGA is reconfigurable logic: a chip whose internal wiring is programmed after manufacture to do exactly the job a system designer needs. In a robot, that job is the unglamorous-but-essential connective tissue — taking the raw outputs of a dozen heterogeneous sensors (cameras, depth sensors, LiDAR, IMUs, force-torque sensors, encoders) and synchronizing, time-stamping, pre-processing and routing them into the main application processor over a clean, low-latency data pipeline. The robot's "brain" (an NVIDIA Jetson or a Qualcomm Dragonwing) is the expensive SoC that runs the model; the Lattice FPGA is the glue layer that feeds it sensor data fast enough and cheaply enough that the brain is not bottlenecked or burdened.

The product families that do this are the Nexus platform — Lattice's low-power FPGA line built on 28nm fully-depleted silicon-on-insulator, including the CrossLink-NX devices optimized for vision and sensor bridging — and the higher-capacity Avant mid-range family. Around the silicon Lattice has built reference solutions: the Holoscan Sensor Bridge, a Lattice-FPGA-based design that creates synchronized, low-latency sensor-data pipelines, is the centerpiece of its robotics push.

Why does a milliwatt-class FPGA become a binding constraint in a robot? Because sensor fusion is a latency-and-power problem that the main SoC is bad at. Offloading depth processing, sensor synchronization and pre-processing onto a dedicated low-power FPGA frees the application processor's compute and power budget for the model — and in a battery-powered humanoid, every watt and every millisecond of latency matters. Lattice's robotics thesis, and the reason this name is the strongest robotics story in this compute-and-materials batch, is that the FPGA is genuinely the right tool for that job, the company has explicitly pivoted to market it that way, and — crucially — Lattice already has a real, growing AI revenue mix in the print, so the robotics angle is incremental upside on a thesis that is already working rather than a pure call option.

Product roadmap

The roadmap has two FPGA platform families and a robotics-specific reference-design layer. On the silicon side: the Nexus platform is the established low-power line — built on 28nm FD-SOI, it spans the CrossLink-NX vision/sensor-bridge devices and the broader small-FPGA portfolio. Lattice extended it with Nexus 2, unveiled December 2024, the next generation of the small-FPGA platform. The mid-range Avant family moved up in capability with the Avant 30 and Avant 50 devices, also introduced December 2024, targeting edge-optimized and advanced-connectivity applications — the parts that handle higher-bandwidth sensor aggregation. Lattice also extended into security with a new secure-control FPGA family carrying crypto-agility and hardware root-of-trust, relevant to robots that need a trusted boot and update path.

The robotics-specific layer is where the 2026 news cluster sits. The Holoscan Sensor Bridge is Lattice's FPGA-based reference design for synchronized sensor pipelines. In March 2026, at NVIDIA GTC, Lattice joined the NVIDIA Halos ecosystem — the physical-AI functional-safety program — committing to build Halos-certified Holoscan Sensor Bridge designs. In April 2026 Lattice announced a collaboration with Texas Instruments to combine TI's sensing technologies with the Lattice Holoscan Sensor Bridge, giving robotics and industrial developers a flexible hardware foundation for synchronized, low-latency sensor pipelines. And at Embedded World 2026 (February 2026) Lattice and Airy3D showcased a humanoid-and-robotic 3D-vision demo combining Airy3D's DepthIQ depth technology with a CrossLink-NX FPGA — depth perception in a very small, low-power form factor, with depth processing offloaded to the FPGA.

What Lattice does not make is the robot brain — it does not make the application SoC or the foundation model, and it explicitly positions itself as the complement to NVIDIA's and TI's silicon rather than a competitor. It also does not make actuators or sensors. The cadence is roughly a two-year platform refresh (Nexus to Nexus 2, Avant capacity steps) layered with a steady stream of reference designs and ecosystem partnerships.

The financial print

Lattice reported Q1 2026 on May 4, 2026: revenue of $170.9 million, up 42% year-on-year and 17% sequentially, with EPS up more than 80% year-on-year to $0.41, above the high end of guidance. The growth driver was the Compute & Communications segment, up 86% year-on-year and 15% sequentially to a record, with management guiding that roughly 38% of 2026 revenue is expected to come from servers and roughly 25% from AI customers. The other half of the business — Industrial & Automotive, which contains robotics — is the steadier, more diversified leg.

For full-year context, Lattice's revenue troughed through a 2024-2025 inventory-correction cycle and is now in a clean recovery — the 42% year-on-year Q1 2026 growth is the inflection. Sell-side coverage is broad for a mid-cap; the analysts who cover the small-FPGA space (Needham, Stifel, Susquehanna, KeyBanc among the historical voices) are generally constructive on the AI-attach and edge-AI mix story, with the debate centered on the valuation. Forward P/E is approximately 53.5x — a growth multiple, but notably less stretched than Ambarella's ~75x, and the technicals are the calmest in this batch: RSI 57.6 is mid-range, +16.5% above the 50-day moving average is a healthy uptrend rather than a parabolic extension. That relative restraint matters — LSCC is the one name in this compute-and-materials group you can initiate without fighting an overbought tape.

The binary event is Q2 2026 earnings, expected early August 2026 (Lattice reports on a roughly quarterly cadence; August 3, 2026 is the working estimate, to be confirmed). The read-through this theme cares about is any incremental disclosure on the Industrial segment and traction on the Holoscan Sensor Bridge robotics design-ins.

Customer mix today

Lattice reports by segment rather than by named customer. The structural shift is the headline: the company has flipped from an industrial-and-consumer-weighted revenue base toward Compute & Communications, which is now the larger and faster-growing segment — up 86% year-on-year in Q1 2026, with roughly 38% of 2026 revenue expected from servers and roughly 25% from AI customers (company guidance, Q1 2026). The other major segment, Industrial & Automotive, is roughly the other half of revenue and is where robotics lives.

The 2024-to-2026 change is the story. In 2024, Lattice was a low-power FPGA company emerging from an inventory correction, with the AI/server attach a promising-but-small contributor. By 2026, AI customers are roughly a quarter of revenue and servers nearly 40% — the AI-attach thesis is no longer speculative, it is in the print. Robotics, sitting inside Industrial & Automotive, is the next leg the company is explicitly building toward, and the cluster of 2026 partnerships — NVIDIA Halos in March, TI in April, the Airy3D humanoid 3D-vision demo in February — is the deliberate effort to convert robotics from a use-case slide into a customer-mix line. Lattice has not yet sized robotics as a revenue percentage, so the honest framing is: the AI/server mix shift is confirmed and in the numbers; the robotics mix shift is in the design-in and partnership stage, expected to show up in the Industrial segment over 2026-2027.

What's actually happening at the NVIDIA Halos and TI partnerships

The mechanism to watch is reference-design adoption. Lattice does not need to win a robot's main SoC socket — it needs the Holoscan Sensor Bridge to become a standard building block in robot designs, the way a particular power-management or connectivity chip becomes a default choice. The two 2026 partnerships are precisely the channels for that. Joining the NVIDIA Halos ecosystem (March 2026, at GTC) means Lattice's Holoscan Sensor Bridge designs get a path to Halos functional-safety certification — and because NVIDIA Jetson is the default robot brain, being the certified sensor-bridge complement to Jetson is a powerful pull-through position: a robot designer building on Jetson now has a Lattice-based, pre-certified sensor pipeline as the path of least resistance.

The TI collaboration (April 2026) is the second channel: combining TI's sensing front-ends with the Lattice Holoscan Sensor Bridge gives industrial and robotics developers a turnkey synchronized sensor pipeline. TI's reach into industrial and robotics design is enormous, so this is distribution leverage. Be specific about what this is and is not: these are partnerships and reference designs, not disclosed dollar-sized contracts or named humanoid production wins. Lattice has not said "robot OEM X has designed in CrossLink-NX for Y million units." What it has is structural positioning — the certified complement to the dominant brain (NVIDIA) and a turnkey design with the dominant industrial-sensing vendor (TI). The qualification-to-revenue conversion runs through 2026-2027 as robot programs that adopt these reference designs move to production. The Airy3D 3D-vision demo is the proof-of-concept that the technical fit is real; the partnerships are the go-to-market; the revenue is the still-to-come part.

The competitive threat / the integrated-SoC squeeze and Microchip/Altera

Lattice's competitive picture has two layers. The direct-competitor layer is the low-power FPGA market itself: Microchip (via its PolarFire and IGLOO low-power FPGA lines) and Altera (the former Intel FPGA business, now independent) are the named competitors for the discrete-FPGA socket. Lattice's defense here is genuine product leadership at the lowest-power, smallest-form-factor end — the Nexus and CrossLink-NX parts are purpose-optimized for exactly the milliwatt-class vision-and-sensor-bridge role a robot needs — and it has historically held strong share at that node. This is a real competitive set but not an existential one; it is a share fight within a category Lattice leads.

The more important layer is the integrated-SoC squeeze. The structural risk is not another FPGA vendor — it is the robot brain itself. If NVIDIA's Jetson and Qualcomm's Dragonwing integrate enough sensor I/O, synchronization and bridging function directly onto the SoC, the discrete low-power FPGA gets designed out of the bill of materials. There is no IP litigation at issue; the threat is roadmap integration. The reason this is a manageable rather than terminal risk — and why Lattice's Halos and TI partnerships are smart — is that Lattice has positioned itself as the complement rather than the competitor: NVIDIA itself blessed the Holoscan Sensor Bridge approach by bringing Lattice into the Halos ecosystem, which is a signal that NVIDIA wants a flexible FPGA glue layer in the design rather than absorbing every sensor function onto Jetson. Lattice's bet is that sensor heterogeneity is permanent — robots will always have a changing mix of sensors that a fixed-function SoC cannot anticipate — and reconfigurable logic is the durable answer to that. That bet is plausible but it is the bet.

The terminal risk

The terminal risk is SoC integration removing the discrete FPGA from the robot. If the robotics-compute industry converges on a small number of highly integrated SoCs that absorb sensor-bridge, synchronization and pre-processing functions on-die, then the low-power FPGA — Lattice's whole franchise in this application — loses its socket. The transition window is the 2027-2030 period as humanoid and AMR bills of materials standardize and SoC vendors push integration to lower system cost. The named beneficiaries of that transition are NVIDIA and Qualcomm, the very SoC vendors Lattice currently partners with.

Lattice's credible defense is structural: sensor heterogeneity and the need for late-stage design flexibility are real and arguably permanent in robotics — a fixed-function SoC cannot anticipate every sensor combination a robot designer will want, and reconfigurable logic is the natural hedge. NVIDIA's decision to bring Lattice into the Halos ecosystem is evidence that even the dominant SoC vendor wants an FPGA glue layer in the architecture. The honest read: this is a slower-burning and more contestable terminal risk than the socket-absorption threat facing a vision-only vendor like Ambarella, because Lattice's function (flexible glue) is harder to fully absorb than a vision block. But it still constrains the multiple — you cannot pay an unlimited price for a company whose role could be integrated away over a five-to-eight-year horizon.

Bull / Gap / Optionality (Photoncap framing)

1. The AI/server mix shift is already in the print — robotics is incremental upside, not the whole thesis. Roughly 25% of 2026 revenue is from AI customers and ~38% from servers (company guidance, Q1 2026), and Compute & Communications grew 86% year-on-year. Unlike the mega-caps where robotics is a 1% rounding error, or the pure-plays where robotics is a pure call option, LSCC has a working AI thesis in the numbers and the robotics angle stacks on top of it.

2. The robotics positioning is the most concrete in this batch. The NVIDIA Halos ecosystem membership (March 2026), the TI sensor-pipeline collaboration (April 2026), and the Airy3D humanoid 3D-vision demo (February 2026) are three distinct, dated, named go-to-market channels — Lattice is the certified sensor-bridge complement to the dominant robot brain and has turnkey distribution through the dominant industrial-sensing vendor.

3. The financial inflection is clean and the technicals are the calmest in the group. Q1 2026 revenue up 42% year-on-year, EPS up over 80%, above the high end of guidance — a clear recovery inflection out of the 2024-2025 inventory correction. And at RSI 57.6 and +16.5% above the 50-day moving average, LSCC is the one name here you can initiate without chasing a parabolic tape.

4. Lattice is the complement, not the competitor, to the robot brain. By positioning the Holoscan Sensor Bridge as the FPGA glue layer that NVIDIA and TI want in the design, Lattice converts the integrated-SoC threat into a partnership — NVIDIA bringing Lattice into Halos is the dominant SoC vendor explicitly endorsing a discrete-FPGA role.

5. The valuation, while a growth multiple, is the most reasonable of the high-multiple names here. A forward P/E of ~53.5x is full but materially below Ambarella's ~75x, Everspin's ~84x or MP's ~49x — and it sits against a confirmed 42% growth rate with the robotics optionality not yet priced in.

Gap

1. Robotics is not yet a sized revenue line. For all three partnerships, Lattice has not disclosed a dollar-sized robotics contract or a named humanoid production design-in. Robotics sits inside the Industrial & Automotive segment as design-ins and reference designs — the conversion to revenue is a 2026-2027 expectation, not a current number.

2. The integrated-SoC squeeze is the structural overhang. If NVIDIA's Jetson and Qualcomm's Dragonwing absorb sensor-bridge function on-die, the discrete low-power FPGA loses its socket over 2027-2030. Lattice's "complement not competitor" positioning is smart, but it depends on SoC vendors choosing not to integrate — a choice they control, not Lattice.

3. The current growth is server/AI-driven, which is its own cycle risk. Compute & Communications up 86% year-on-year is great, but it ties a large and growing share of revenue to the AI-server capex cycle — if hyperscaler FPGA-attach spending cools, the headline growth rate decelerates hard, independent of anything happening in robotics.

4. The FPGA competitive set is real. Microchip (PolarFire, IGLOO) and the independent Altera both compete for the low-power FPGA socket. Lattice leads at the lowest-power node, but it is a share fight, and pricing pressure in a contested category caps margin expansion.

Optionality

The trade

LSCC is a Bucket B name and the strongest robotics-thesis name in this compute-and-materials batch: a low-power FPGA franchise with a confirmed AI/server mix shift already in the print, the most concrete robotics go-to-market positioning of the group, and — uniquely here — a calm, non-extended tape. Initiate in a $117-130 entry zone (current $123.63 minus roughly 1xATR to plus 5%), size at 1.5-2.5% of risk capital — the cleaner technicals and the in-the-print AI thesis support a fuller position than the extended names — with a stop near $104 (below the 50-day moving average and the prior consolidation shelf). The defining near-term binary is Q2 2026 earnings, expected around August 3, 2026, where the read-through is the AI-attach growth rate and any Industrial-segment robotics commentary. If you want a higher-beta expression of the same physical-AI sensor thesis you take the SoC names (NVDA, QCOM); if you want the dedicated-vision angle, Ambarella (AMBA) — but LSCC is arguably the cleanest risk/reward in the batch because the base thesis already works and robotics is free optionality on top. LSCC's role in a robotics book is a core position, not a satellite — the one name here where the entry tape, the valuation and the thesis-clarity all line up. Conviction: 7 / 10.

Sources referenced inline throughout. Reference v1 of this template format: _Watchlist/hanmi-photoncap-style.md.

\newpage

NVDA — NVIDIA Corporation · BUY (Tier-1) · Conv 7/10 · Bucket B

ticker: NVDA name: NVIDIA Corporation theme: Robotics bucket: B conviction: 7 entryzonelo: 221.13 entryzonehi: 244.41 currentprice: 232.77 pricedate: 2026-05-14 positionsizepct: 2.5 stoploss: 198.00 thesisoneline: The default compute and software stack for general-purpose robotics — Jetson Thor brains plus Isaac/GR00T/Omniverse — though robotics is still ~1% of a $216B revenue base. catalystnext: Q1 FY27 earnings catalystdate: 2026-05-20 deepdivepath: Theme -- Robotics/NVDA/nvda-deep-dive.md lastupdated: 2026-05-14T00:00:00Z rsi: 75.5 vs50ma: 21.1 forwardpe: 20.5 themecycleposition: early customermixsummary: Hyperscalers ~50%+ of data center; robotics customers (Figure, Boston Dynamics, Amazon, Caterpillar, Agility) are design wins not yet material revenue. terminalriskoneline: Custom ASICs (hyperscaler in-house silicon) and a fragmenting robotics-compute field could erode the GPU/CUDA moat before robotics revenue scales enough to matter. bulldriverscount: 5 gapriskscount: 4 optionalitycount: 6 lastearningsdate: 2026-02-25 nextearningsdate: 2026-05-20