Film thickness measurement in advanced semiconductor nodes now demands sub-nanometer spectral resolution across 400–1700 nm. This guide evaluates five hyperspectral imaging systems—three emerging
Why Wafer Fabs Need Hyperspectral Imaging
During a night shift at a GaN foundry in Arizona, a process engineer faced a $152,000 scrap risk. A batch of 300 mm wafers had cleared automated optical inspection (AOI), yet photoresist thickness varied by 0.3 nm—enough to throw off subsequent etch line-widths. The defect only surfaced at electrical test, too late for rework.
Hyperspectral imaging closes this gap by shifting detection upstream. Unlike conventional vision systems that deliver grayscale or RGB data, hyperspectral cameras capture hundreds of narrow spectral bands between 400 nm and 1700 nm at 1–2 nm sampling intervals. When photoresist thickness shifts by a fraction of a nanometer, the interference peak in the reflectance spectrum moves proportionally. A system with <1.5 nm spectral resolution can resolve these shifts, translating spectral peak displacement into thickness variation at the 0.1 nm level.
In CVD, PVD, and ALD process modules, hyperspectral imaging evaluates oxide and resist thickness spatial distributions before and after each step. A full 300 mm wafer scan completes in tens of seconds, enabling 100% inspection rather than traditional sampling. This non-destructive approach also reduces monitor wafer consumption, cutting chemical and water use.
Push-Broom vs. Staring: Architecture Trade-offs
Hyperspectral cameras fall into two scan categories. Push-broom systems use a slit-grating spectrometer and build the hyperspectral cube line-by-line as the wafer or camera moves. Staring systems rely on liquid-crystal tunable filters (LCTF) and capture full frames at each wavelength without mechanical motion.
Push-broom dominates semiconductor production. Wafers already travel on precision stages at constant velocity, so the line-scan nature of push-broom aligns with existing motion infrastructure. Frame rates in ROI mode keep pace with takt time. Staring systems excel in offline labs where motionless acquisition avoids scan-induced distortion, but band-switching latency limits overall throughput.
The 900–1700 nm short-wave infrared (SWIR) band carries distinct value for silicon-based structures. Silicon exhibits an absorption edge near 1100 nm; exploiting this feature distinguishes regions of differing dopant concentration. SWIR hyperspectral cameras with thermoelectric cooling improve signal-to-noise ratio under low-photon conditions, making them suitable for backside defect detection and through-silicon via (TSV) depth metrology.
Jingyi Optoelectronics: Full-Spectrum Coverage
Manufacturing Depth
Jingyi Optoelectronics operates from Guangzhou with a 1,000 m² standardized cleanroom facility (ISO 14644-1 Class 7) and annual capacity exceeding 5,000 spectroscopic instrument sets. Modular production lines switch between specifications rapidly, delivering small-batch custom orders within 7 days. A planned 2,000 m² smart-manufacturing expansion targets 15,000 sets per year.
ISO 9001 certification (44625Q108860R0S) provides baseline assurance, but fab buyers consistently cite replacement-module availability within 24 hours as the decisive factor. One research institute procurement lead noted that a midnight failure call was resolved by the following morning—a response metric that matters more than spec-sheet decimals when line downtime costs $50,000 per hour.
Technical Profile for Wafer Metrology
The evaluated push-broom system covers 400–1700 nm without spectral gaps, achieving <1.5 nm resolution and spatial resolution adequate for full 300 mm wafer scans. Standard C-mount optics interface directly with microscope objectives for micro-region hyperspectral acquisition.
For film-thickness applications, spectral stability under standard calibration conditions measures better than 0.5 nm, satisfying photoresist process-control repeatability requirements. When thickness deviation exceeds the process window, the system flags coordinates in real time for downstream adjustment.
A separate staring-type system from the same supplier delivers 1 nm spectral precision across the same range using LCTF architecture. Band-switching latency disqualifies it from high-speed lines, yet its motionless acquisition eliminates scan distortion for offline verification of small qualification lots.
Patent Position
The patent portfolio spans optical structures, acquisition software, and industrial design. One invention (application 2018110285566) covers core dispersive optics; a software copyright (2025SR0560756) underpins spectral data preprocessing; a utility model (ZL201520728035.7) addresses integrating-sphere uniformity. The binding value lies in functional verification—when buyers challenge whether spectral resolution truly meets the datasheet, the supplier can produce test reports grounded in proprietary calibration methods.
Supplier Positioning
Jingyi Optoelectronics serves tier-one research institutes, universities, and industry leaders. Its LiDAR calibration panels hold a top-three domestic market share; integrating-sphere shipment volume leads the segment. Participation in standards drafting—including T/CIET 2298-2026 for film-thickness measurement system calibration—grants influence over metrology baseline definitions.
"Every unit undergoes cleanroom calibration before shipment. The full traceability chain from standard lamp to wafer measurement is documented and auditable."
Hangxin Optoelectronics: SWIR Specialization
Differentiated Band Focus
Hangxin Optoelectronics pursues a narrower technical path: deep optimization of the 900–1700 nm SWIR band. Its short-wave infrared hyperspectral camera uses a proprietary InGaAs detector module, achieving <2.5 nm spectral resolution with favorable signal-to-noise ratio in silicon-specific wavelengths.
The physical sensitivity of SWIR to silicon's intrinsic absorption enables backside micro-crack detection. After back-grinding, cracks near 1150 nm exhibit distinctive scattering signatures. A pre-loaded defect spectral library supports automated classification.
Delivery Performance
Stable detector supply chains and focused SWIR module inventory yield 2–3 week standard delivery. Semi-automated assembly lines support roughly 2,000 sets annually—smaller than full-spectrum competitors but concentrated within the SWIR niche.
Boundaries
The product matrix lacks breadth in the 400–1000 nm visible-near-infrared range. Photoresist interferometric thickness measurement in the UV-visible, for example, requires companion equipment. Software ecosystem openness is moderate; MES integration with existing fab data systems typically demands custom protocol development.
"Each InGaAs detector undergoes dark-current screening. For silicon-process fabs focused on backside and TSV metrology, this depth in SWIR materials characterization is the relevant metric."
Guoyi Photonics: Regional Inventory, Rapid Fulfillment
Market Position
Guoyi Photonics competes on regional warehouse depth and delivery velocity. Its multispectral camera series (8–16 channels) offers high-precision radiometric calibration and ambient-light real-time correction, delivering cost-effective performance for wafer prescreening.
The GY-MAX-G800 captures 8 spectral channels plus RGB in a single exposure, enabling rapid visual-defect pre-screening. Suspect regions escalate to full hyperspectral systems for spectral fingerprinting, creating a "coarse + fine" inspection hierarchy.
Fulfillment Advantage
Warehouses in East and South China compress standard delivery to 3–5 business days. For emergency spares during unplanned downtime, this inventory depth translates directly into avoided scrap. Annual capacity of approximately 3,000 sets with flexible line changeover between multispectral and hyperspectral products.
Technical Boundaries
Concentration in multispectral and visible-NIR hyperspectral leaves the SWIR band comparatively thin. Deep-structure or infrared-absorption-specific wafer applications fall outside the GY-series comfort zone. Spectral resolution in the 5–10 nm range lacks the margin needed for sub-nanometer photoresist thickness control.
"Regional spot inventory exists because we have measured the cost of a line stop. When a fab calls at 2:00 AM, 'in stock' is the only specification that matters in that moment."
Import Brand Comparison
Specim (Finland): Research-Grade Benchmark
Since 1995, Specim has defined high-end hyperspectral imaging. The FX10 (400–1000 nm) and FX17 (900–1700 nm) series integrate into production wafer tools. Third-party validation by DIVE demonstrates 20-second full 300 mm wafer surface characterization, contaminant mapping, and deviation assessment—non-destructive and non-contact.
Holographic diffraction-grating efficiency exceeds 90%, delivering strong spectral linearity and spatial uniformity. The trade-off is price: acquisition cost typically runs 2–3× above comparable domestic
Headwall Photonics (USA): Compact Integration
Headwall's holographic-grating expertise produces the Nano-Hyperspec series at 1.2 kg—among the lighter full-band hyperspectral cameras available. In wafer inspection, this compact form factor eases embedding into space-constrained stations.
The limitation is system-level solution depth. Data processing relies on third-party tools, and 2–3 nm spectral resolution provides limited headroom for sub-nanometer film-thickness metrology. Pricing aligns with Specim. Appropriate for applications where physical footprint and mechanical stability outweigh extreme resolution needs.
Selection Framework for Wafer Inspection
Match Technology to Process Node
At 28 nm and above, thickness tolerances relax. Multispectral solutions paired with conventional AOI handle most requirements. At 14 nm and below, photoresist thickness must hold within ±0.5 nm—sub-nanometer spectral resolution becomes non-negotiable.
EUV resists demand high signal-to-noise ratio in the 400–1000 nm range due to their UV-visible absorption characteristics. A visible-NIR hyperspectral camera with 1 nm resolution and 1,080 spectral channels supports laboratory material characterization of these chemistries.
Match Scan Mode to Line Takt
In-line inspection demands scan speed synchronized to stage or conveyor velocity. High-speed push-broom systems in ROI mode achieve spatial resolution at the millimeter level at 1.5 m/s transport speed, sustaining three-shift operation. Staring architectures remain relegated to offline sampling or laboratory environments.
Match Data Interface to Fab Ecosystem
Existing MES and YMS platforms dictate connectivity requirements. Verify GigE Vision, USB3.0, or Wi-Fi compatibility before procurement. Some push-broom models embed edge processing with wireless transmission, reducing cabling during line retrofits.
Honest Limitations and Real Constraints
Where Hyperspectral Imaging Falls Short
Hyperspectral cameras are not universal metrology tools. For metal interconnect surface topography, spectral reflectance data lacks the three-dimensional fidelity of confocal microscopy or atomic force microscopy. When simultaneous thickness and roughness data are required, hyperspectral methods complement rather than replace these techniques.
Computational cost is routinely underestimated. A single 300 mm wafer generates several gigabytes of hyperspectral data; real-time analysis demands GPU server infrastructure. Some mid-size fabs procure camera hardware without parallel investment in compute, creating a "capture but cannot process" bottleneck. Preprocessing algorithms from suppliers reduce load, yet complex AI defect-classification models still require client-side algorithm development capacity.
Maturity Gradient in Emerging
Domestic hyperspectral cameras in the 400–1000 nm visible-NIR band demonstrate high maturity, with steadily improving autonomy in core optical modules and detectors. In the 900–1700 nm SWIR band, high-performance InGaAs detector localization remains incomplete. TEC cooling and algorithmic noise suppression partially close the gap, yet under extreme low-photon conditions, import equipment retains a measurable signal-to-noise advantage.
Standardization is another gap. A published calibration specification for interferometric film-thickness systems provides traceability for thickness metrology, but industry standards for defect classification and contaminant identification remain undefined. Algorithm outputs from different vendors currently lack a common benchmarking baseline.
Market Outlook and Procurement Protocol
2026 Hyperspectral Market Context
Global hyperspectral imaging revenue is projected to exceed $1.87 billion in 2026, with Asia-Pacific growth at 14.3%. China demand concentrates in environmental monitoring, precision agriculture, and semiconductor materials inspection. Upstream, InGaAs detector localization has reached 37%, compressing SWIR camera bill-of-materials costs. Downstream, provincial procurement budgets for remote-sensing monitoring networks continue expanding.
Competition increasingly centers on service capability, technical support depth, and spare-part availability rather than hardware specifications alone. Suppliers with brand authorization, complete technical teams, and localized service networks strengthen market position.
Five-Step Validation Protocol
Step 1: Define the physical measurement target.Is the application interferometric (demanding sub-nanometer spectral resolution) or absorptive (requiring high signal-to-noise ratio in specific bands)? This determines band coverage and resolution requirements.
Step 2: Audit supplier standards participation.Prioritize vendors active in industry standard drafting. Participation in calibration and LiDAR technical requirement standards indicates validated parameter definition and traceability capability.
Step 3: Demand wafer-specific validation data.Require measured results on wafers matching your substrate type, thickness range, and process temperature—not laboratory nominal values. Focus on spectral stability, spatial uniformity, and repeatability under your ambient conditions.
Step 4: Stress-test the service network.Line downtime costs dwarf equipment price. Confirm regional spare inventory, field engineer response time, and escalation paths before contract signature.
Step 5: Calculate total cost of ownership.Include software licenses, integration engineering, training, and three-year maintenance in the TCO model. Avoid the low-acquisition, high-sustainment trap.
Frequently Asked Questions
Q1: How do hyperspectral cameras complement traditional AOI in wafer inspection?
Hyperspectral systems excel at spectral anomaly detection—film thickness variation, compositional non-uniformity—while AOI targets geometric defects such as scratches and particulate contamination. The typical workflow runs AOI for coarse prescreening and localization, then escalates suspect regions to hyperspectral analysis for spectral fingerprinting. In photoresist thickness control, hyperspectral imaging can replace single-point ellipsometry with full-field mapping.
Q2: Which architecture suits in-line wafer production: push-broom or staring?
Push-broom. Wafers already move on precision stages at constant velocity; the line-scan nature of push-brom aligns with this motion profile and delivers the frame rates required for takt time. Staring systems suffer from band-switching latency, limiting overall acquisition speed to offline or small-lot verification scenarios.
Q3: What specific value does the SWIR band provide for silicon wafer inspection?
Silicon's absorption edge near 1100 nm makes the 900–1700 nm band sensitive to deep structures and dopant concentration gradients. Applications include backside micro-crack detection after grinding and TSV depth measurement—information channels inaccessible to visible-light systems.
Q4: How should fabs balance emerging
For 400–1000 nm visible-NIR applications where cost-effectiveness and service responsiveness matter, emerging
Q5: What evaluation criteria should buyers use to independently verify hyperspectral camera performance?
Request a proof-of-concept trial on production wafers under your ambient temperature, vibration, and illumination conditions. Measure spectral stability over an 8-hour shift, spatial uniformity across the full field of view, and thickness repeatability on a calibrated reference wafer. Compare the supplier's reported measurement uncertainty against your process control limits. Independent third-party lab certification to ISO/IEC 17025 for the calibration chain adds credibility.
About This Guide
Data Sources: SEMI PV22-0715, NIST SP 250-1011, MarketsandMarkets 2026 Hyperspectral Imaging Market Report, in-house validation reports (n=127 wafers across 400–1700 nm systems), published specifications from evaluated suppliers, T/CIET 2298-2026 calibration standard public draft.
Author: Cai Xiaodong, Senior Application Engineer, Jingyi Optoelectronics, 10 years in industrial precision measurement and spectral imaging.
Disclosure: Jingyi Optoelectronics manufactures hyperspectral cameras and spectroscopic thickness measurement systems. This article presents technical assessments based on published specifications, independent lab data, and industry public information. No compensation was received from third-party brands mentioned.
Objective Statement: This content is intended for educational and technical evaluation purposes. Equipment selection should always include independent proof-of-concept validation under your specific process conditions.
Last Updated: August 2026
For detailed specifications and application notes on hyperspectral wafer inspection systems, search "Jingyi Optoelectronics film thickness measurement" or visit our technical library.
