Fast Fiber-to-Chip Peak Search for Wafer- Level Testing of Photonic Integrated Circuits

Fast Fiber-to-Chip Peak Search for Wafer- Level Testing of Photonic Integrated Circuits
  • 所属分类:应用中心
  • 产品归属:
  • 产品概况:
  • Sub-second open-loop spiral scanning with position-referenced power mapping using the TRITOR 100 SG three-axis piezo stage and the NV 40/3 CLE controller


    Abstract

    Optical input/output testing is the throughput bottleneck of wafer-level photonic integrated circuit (PIC) manufacturing: every device under test first requires active sub-micrometer alignment of an optical probe before any measurement can begin. This application note describes a production wafer-test system in which two TRITOR 100 SG three-axis piezo nanopositioning stages, each driven by an NV 40/3 CLE three-channel controller, perform the fast fiber-to-chip peak search on the input and output side of the device under test. An externally generated spiral trajectory is applied through the analog modulation input of the controller, while the strain-gauge position monitor output is sampled synchronously with the photodiode power signal. The result is a position-referenced coupling map acquired in a single open-loop scan of approximately 0.8 s, contributing to a total per-site alignment time of approximately 5 s – compatible with passive-device test cycles of around 10 s and full-wafer optical test runs measured in hours rather than days.

    1. Background: From Electrons to Photons – and Why Test Throughput Decides Who Scales

    Data center interconnects are migrating from electrical to optical signaling. Silicon photonics and co-packaged optics (CPO) move the optical interface from the faceplate of the switch directly to the package of the ASIC, promising higher bandwidth density and lower energy per bit for AI and hyperscale workloads. Market analysts consistently project annual growth rates of 25–30% for silicon photonics and co-packaged optics over the next decade, driven primarily by AI data center infrastructure.


    This transition has a manufacturing consequence that is easy to underestimate: photonic chips must be tested optically, at wafer level, before dicing and packaging. Unlike electronic die, a photonic die cannot be qualified through electrical probing alone– light must be coupled into and out of on-chip waveguides for every device under test. Because co-packaged assemblies have very limited rework ability, known-good-die assurance at wafer level is not optional; a single untested bad die can scrap an entire multi-die package.


    The industry consensus is equally clear about where the cost sits: optical alignment is the dominant time and cost driver in photonic test and packaging, with packaging and alignment commonly cited as up to 80% of total photonic module cost. Every photonic wafer prober, every test house and every PIC manufacturer therefore faces the same core problem – how to find the optical coupling maximum between a fiber probe and an on-chip coupler in seconds, thousands of times per wafer, around the clock.

    2. The Technical Challenge

    2.1 Sub-micrometer tolerances, no mechanical reference

    A single-mode fiber has a field diameter of roughly 10 μm; onchip edge couplers and grating couplers exhibit 1-dB alignment tolerances on the order of ±0.5 to ±1 μm, and significantly tighter for high-efficiency spot-size converters. No machine-vision or mechanical reference can place a probe within these tolerances repeatably. The only valid reference is the optical power itself – alignment must be performed actively, by scanning the probe and maximizing the measured coupling.

    2.2 The throughput arithmetic

    The arithmetic is unforgiving. A wafer-level optical-optical (OO) test of a fully populated wafer can occupy a prober for six to seven hours even when each passive device is tested in approximately ten seconds. Within that ten-second budget, the optical alignment – typically performed independently on the input and the output side of the device – must not consume more than a few seconds, including coarse capture and fine peak search on both probes. Every additional second of alignment time per site adds roughly an hour of prober time per wafer at 3,000+ test sites.

    2.3 The control dilemma: speed versus position truth

    Piezo actuators are the only actuator class fast and fine enough for this task, but they confront the system designer with a dilemma:

    • Open-loop operation exploits the full mechanical bandwidth of the stage and allows scan trajectories at tens of hertz – but piezo hysteresis (typically 10–15% of travel) and creep mean the commanded voltage does not correspond linearly to the actual position. A power map plotted against the command signal is geometrically distorted, and the located “peak” may be offset from the true coupling maximum.
    • Closed-loop operation linearizes the motion via sensor feedback, but the servo loop – sensor noise filtering, controller bandwidth, stability margins and settling behavior – limits usable trajectory frequency and adds phase lag. For a fast spiral or raster scan, a conservative closed-loop configuration can cost a significant share of the achievable scan speed.

    Most commercial “alignment engine” products resolve this dilemma inside a closed, proprietary controller. That is a valid approach – but it forces the equipment builder to adopt the vendor’s controller ecosystem, algorithms and software stack, and it makes the alignment module a black box inside the tool.

    3. The Application: Dual-Probe Peak Search Inside an Automated Wafer Prober

    A manufacturer of photonic wafer-test equipment integrated PSJ piezo systems into an automated prober for optical-optical and optical-electrical chip testing. The architecture separates coarse and fine positioning cleanly:

    • The wafer stage steps from site to site according to the wafer map; machine vision provides coarse mechanical positioning of the optical probes.
    • Two TRITOR 100 SG stages – one carrying the input fiber probe, one the output probe – perform the fine alignment, each driven by an NV 40/3 CLE three-channel controller.
    • A tunable laser feeds the input probe; the transmitted light is detected by a photodiode whose current is converted by a logarithmic transimpedance converter covering 0.1 nA to 10 mA – eight decades of optical dynamic range in a single analog signal.
    • A multifunction data-acquisition (DAQ) card orchestrates the scan: its analog outputs generate spiral voltage waveforms applied to the modulation (MOD) inputs of the NV 40/3 CLE; its analog inputs simultaneously sample the converted optical power and the controllers’ position monitor (MON) outputs.

    3.1 The key idea: open-loop speed, sensor-referenced data

    The decisive architectural element is the synchronous acquisition of the Monitor signal. The strain-gauge sensors integrated in the TRITOR 100 SG measure the actual motion of the stage platform; the NV 40/3 CLE makes this measurement available as a real-time analog output. The system therefore scans open-loop – at the full dynamic capability of the stage, unconstrained by servo bandwidth – while the coupling map is constructed from measured positions, not commanded voltages.

    Hysteresis and creep still occur in the motion, but they no longer corrupt the data: each power sample is paired with the position at which it was actually acquired. The peak of the resulting three-dimensional power map is the true coupling maximum, and the stage can subsequently be commanded to it. The approach combines the speed advantage of open-loop scanning with the position fidelity of a sensor-based system – without requiring the servo loop to track the fast trajectory.

    Figure 1: Spiral fine-scan principle – in a single 0.8 s open-loop pass; the coupling maximum is located from the position-referenced map (schematic representation).
    Figure 2: (a) Modulation input command vs. Monitor Output strain-gauge response at the 26 Hz operating frequency – phase lag and amplitude deviation are visible but irrelevant to the map, since power and position share one time base. (b) Open-loop hysteresis made observable through the MON output (schematic representation).

    3.2 Measured performance

    Parameter Value
    Open-loop spiral scan time approx. 0.8 s
    Scan trajectory frequency 26 Hz
    Alignment sequence per site coarse + fine scan, input and output probe
    Total alignment time per test site approx. 5 s
    Resulting passive-device test cycle approx. 10 s
    Full-wafer OO test 6-7 h
    Table 1: Alignment timing achieved in the production system (customer data).
    Figure 3: Sequential dual-sided alignment within the 5 s budget. Fine-scan durations (0.8 s) are measured values; the split of the remaining time between the coarse phases is shown schematically.
    Figure 4: Resulting coupling map around the optimum with 0.5 / 1 / 3 dB excessloss contours – the sub-micrometer 1 dB tolerance window illustrates why active alignment is mandatory (schematic representation).

    4. Engineering Difficulties – and How They Were Addressed

    • Hysteresis and creep in open loop. Addressed at the data level via synchronous MON sampling (Section 3.1) rather than at the actuation level. The strain-gauge measurement system of the TRITOR 100 SG was specified precisely for this: it makes hysteresis observable instead of attempting to suppress it during the fast scan.
    • Phase lag between command and response. At 26 Hz the stage response trails the command waveform in phase and amplitude. Because power and MON position are sampled on simultaneous DAQ channels, this lag is irrelevant to the map – both signals share the same time base. The customer’s dual-channel oscilloscope verification (MOD command vs. MON response) confirmed stable, repeatable transfer behavior at the operating frequency.
    • Trajectory frequency versus mechanical resonance. The TRITOR 100 SG offers resonant frequencies of approximately 500 Hz per axis. Operating the spiral at 26 Hz keeps the excitation more than an order of magnitude below resonance, avoiding ringing and trajectory distortion while leaving headroom for faster scan profiles.
    • Optical dynamic range. During first-light capture the coupled power can be many decades below the optimum. The logarithmic converter (0.1 nA–10 mA) keeps the entire search range within one ADC input span – no gain switching during the scan.
    • Dual-sided alignment. Input and output probes interact: optimizing one side shifts the apparent optimum of the other. The sequential coarse/fine strategy on both sides (left coarse → left fine → right coarse → right fine) converges reliably within the 5 s budget.
    • 24/7 production stability. Solid-state flexure guidance in the TRITOR 100 SG involves no friction, no lubricants, no particle contamination and no mechanical play – there is no wear mechanism in continuous scanning operation, a prerequisite for prober uptime.

    5. The PSJ Solution

    5.1 TRITOR 100 SG – three-axis piezo nanopositioning stage

    Property TRITOR 100 SG
    Axes 3 (X, Y, Z), parallel-kinematic flexure
    design
    Travel 100 μm per axis open loop / 80 μm
    closed loop
    Resolution 0.2 nm open loop / 2 nm closed loop
    (typ., with NV 40/3 CLE)
    Resonant frequency approx. 500 / 550 / 480 Hz (X/Y/Z)
    Integrated sensor Strain gauge (SG) per axis
    Dimensions 40 × 40 × 34 mm³
    Guidance Solid-state flexure hinges – frictionless,
    zero play, no wear

    The 100 μm three-axis scan volume comfortably covers the capture range required after vision-based coarse positioning, while the compact 40 mm cube integrates into the confined probe-head space above a wafer chuck – alongside electrical probes, cameras and a second optical probe.


    5.2 NV 40/3 CLE – three-channel piezo controller

    • Three amplifier channels in one bench-top unit – one controller per three-axis probe.
    • Analog MOD input per channel: external waveform generation (DAQ, FPGA, function generator) drives the stage directly. The alignment algorithm remains the property and design choice of the equipment builder.
    • Analog MON output per channel: real-time strain-gauge position signal for synchronous acquisition – the enabler of position-referenced open-loop mapping.
    • Closed-loop operation available on all channels where linearized point-to-point positioning is preferred over scan speed – selectable per process step.
    • USB/RS-232 interface for set-point control and configuration from the tool software.

    5.3 Why an open architecture matters here

    Integrated alignment engines from larger motion vendors deliver impressive turnkey algorithm performance, with fine-alignment routines completing in well under a second. For equipment builders who want a sealed subsystem, they are a legitimate choice. The PSJ approach addresses a different – and in our experience growing – customer profile: photonic test equipment manufacturers who consider the alignment strategy part of their own intellectual property and competitive differentiation. For them, PSJ supplies the precision mechanics and the analog real-time interfaces; the customer owns the algorithm, the timing, the software stack and the system integration. This division of labor avoids controller lock-in, keeps the bill of materials lean, and lets the alignment routine evolve with the customer’s test recipes instead of a vendor’s firmware roadmap.

    6. Beyond Chip Testing: Where This Architecture Scales

    The same fast peak-search building block – compact three-axis piezo stage, analog drive, synchronous position read-back – applies wherever light must be coupled actively between a fiber and a small aperture:

    • Wafer-level and die-level test of PICs for transceivers, CPO and optical I/O – the application described here.
    • Coupling light onto photodiodes and detector arrays for characterization of optical-electrical devices.
    • Active alignment in photonic packaging: fiber attach, lens attach, laser-to-PIC coupling prior to bonding or welding.
    • Burn-in and life-test stations where coupling must be reoptimized periodically to track thermal drift.
    • Quantum photonics, LiDAR and sensing modules with similar fiber-to-chip interfaces.

    As co-packaged optics moves toward volume production in the second half of this decade, the number of optical I/Os per wafer – and therefore the number of alignments per wafer – rises steeply. Test capacity scales with alignment speed. Equipment builders who solve the alignment block early, with an architecture they control, hold a structural cost advantage.

    7. Summary

    Two TRITOR 100 SG stages with NV 40/3 CLE controllers perform the complete dual-sided fiber-to-chip peak search of an automated photonic wafer prober in approximately five seconds per test site, using externally generated 26 Hz spiral scans completed in 0.8s. The combination of open-loop scan speed and synchronous strain-gauge position read-back delivers geometrically correct coupling maps without the bandwidth penalty of closed-loop trajectory tracking. The open analog architecture leaves algorithm and system intelligence with the equipment builder – a deliberate alternative to closed, proprietary alignment engines.


    If you are developing photonic test, packaging or alignment equipment and want to evaluate this architecture against your cycle-time targets, our application engineers will support you with stages, controllers, reference circuitry and measured step-response and frequency-response data for your specific drive configuration.

在线咨询