FLIR Boson Integration Guide: Pinouts, Sync Setup & OEM Module Alternatives

Integrating an uncooled Long-Wave Infrared (LWIR) core into an inspection drone, robotic rover or embedded camera requires a configuration-specific design. The FLIR Boson family uses 12 µm uncooled Vanadium Oxide (VOx) technology; match its connector, video modes and synchronization settings to the exact part number, firmware and carrier.

This guide covers pinouts, synchronization, Linux capture, optical planning and procurement for civilian inspection and robotics. It compares Boson with AeroMini 640’s interface-kit route and SuperMini 640/640T’s compact bare-core architecture. Changing families requires an electrical, mechanical and software review; pin compatibility is not established here.

Documentation note, 6 October 2026: Freeze the selected model, lens, interface board and firmware before design release.

Technical author: Daniel · Hardware Support
Sales contributors: Vivian, Lena and Sophie

1. Electrical Interface, High-Density Pinouts & Carrier Board Design

The FLIR Boson Product Data Sheet, revision 380, section 4 describes the 80-pin Hirose DF40C-80DP-0.4V(51) interface for the documented Boson core. Use its pin table and connector drawing together: connector numbering, mating direction, return paths and signal voltage matter as much as the connector family. An evaluation carrier’s external socket can have different pin assignments and power requirements from the bare core.

Separate supply voltage from logic voltage. Revision 380, Table 23 specifies a 3.14–3.46 V core supply, nominally 3.30 V, with a listed 60 mV peak-to-peak ripple maximum. Its bare-core EXT_SYNC signal is in the 1.8 V domain (Table 3). A 3.3 V supply specification is not permission to drive a 3.3 V trigger into that signal. Review the revision-matched electrical limits and any level translation on the carrier before connecting an MCU, FPGA or host GPIO.

For a new PCB, identify each bus before routing. USB differential pairs, MIPI lanes and parallel video have different electrical and timing requirements. Follow the camera and receiver design guides, then have the PCB manufacturer calculate the impedance for the chosen stackup. Keep high-speed return paths continuous and separate camera supplies and routing from motor drivers and switching nodes. Avoid applying a universal trace-skew or ripple limit across unrelated cores.

Power-delivery validation belongs on the bench as well as in the schematic review:

• Check the actual input: Distinguish bare-core rails from an interface board’s supply input. Confirm allowed voltage, startup current and power-on sequencing before applying power.
• Measure at the camera: Check voltage, transients and noise at its connector under startup, steady capture, correction events and representative host activity. Use the limits in the selected core’s manual.
• Budget regulation and heat: Select filtering and decoupling for the load. Check regulator dropout, dissipation and frequency-dependent noise rejection.
• Test the installed system: Repeat capture with motors, radios and processors operating. Record artifacts, dropped frames and recovery behavior.

2. Digital Video Transport: MIPI CSI-2, USB UVC, and Industrial Standards

Choose video transport from the required data, host receiver, cable route and measured timing. An interface name alone does not specify the available pixel formats or the delay from the scene to the application.

USB UVC: Boson documentation includes USB video support, and FLIR’s software-version table identifies USB 3.0 support from release 2.1; earlier releases support USB 2.0. Check the actual camera version and carrier rather than labeling the entire family USB 2.0-only or USB 3.0-only. On a Linux host, enumerate the camera’s advertised formats, sizes and intervals before selecting a stream. V4L2 Y16 is 16-bit little-endian grayscale storage, and the actual sampling precision can be lower. Its values may represent pre-AGC image counts or a documented radiometric output, depending on configuration; the storage format alone does not prove calibrated temperature data.

MIPI CSI-2 and parallel video: MIPI is available on particular Boson configurations; confirm the part number, firmware and supported output mode against the matched documentation. A carrier connector is not proof of host support. The receiver still needs correct lane mapping, clocking, packet format and driver configuration. Parallel video likewise requires the documented data and synchronization timing. Neither path guarantees sub-millisecond end-to-end latency.

Industrial transport: Longer camera-to-processor runs may require a bridge or camera assembly. Camera Link, Ethernet and serialized links need matched transmitters, receivers, cables and software; they are not native outputs of every core. Validate cable distance and frame integrity on the complete link.

For each candidate, record three separate deliverables: displayable imagery, radiometric data if required, and command/control access. Validate that the chosen board can provide the required combination at the selected rate; a colorized CVBS picture is not a per-pixel temperature stream.

3. Hardware Synchronization (EXT_SYNC) for Stereo Thermal & Multi-Sensor Fusion

Thermal stereo and multi-sensor inspection systems need a clear timing model. Software arrival timestamps include transport and host scheduling effects; hardware synchronization can improve alignment when both cameras and the surrounding electronics support it. It does not remove the detector’s response time, image-processing delay or clock-calibration requirements.

Start with the Boson EXT_SYNC electrical definition. The bare-core interface in revision 380, Table 3 uses 1.8 V logic. Confirm whether the selected carrier exposes that signal directly or through a translator. Use a suitable electrical interface for the documented signal; do not drive a bare-core 1.8 V EXT_SYNC pin directly from a 3.3 V GPIO.

• Select a documented synchronization mode: Revision 380, section 6.17 describes Master/Slave synchronization from R2.1 for fast configurations. Slave mode responds to a rising edge. Confirm that the ordered camera supports the mode, and check which device drives the line; never connect two active outputs together.
• Match the timing specification: The same section recommends a pulse width greater than 90 ns and describes nominal 60 Hz sensor timing. Averaging can produce 30 Hz output from 60 Hz acquisition. These details do not replace its complete timing requirements: review clock continuity, the supported input cadence and CMOS one-shot versus continuous output for the selected configuration.
• Measure scene-to-frame behavior: Observe synchronization signals and frames during a repeatable scene change. Distinguish acquisition, internal processing, USB transfer and application arrival.
• Handle correction events: Record flat-field correction (FFC/NUC) behavior, duplicate or interrupted frames, and the application’s recovery path. Carry frame status into the fusion pipeline rather than treating every delivered frame as a new observation.

Keep detector response and pipeline delay separate. Revision 380, section 5.10 gives original Boson pipeline examples of approximately 24 ms pre-AGC, 25 ms post-AGC and 43 ms post-zoom, while Boson+ pre-AGC is below 6 ms. Its nominal 8 ms thermal time constant is a separate detector property. These are documented camera-stage figures, not guarantees for USB transport, Linux, ROS 2, display or inference latency. Use them to plan measurements, not as one fixed correction that makes a software timestamp exact.

Validate geometric calibration, frame timing and host-clock alignment over the expected temperatures and motion. A sync pin or ROS 2 image topic alone does not establish sub-millisecond sensor alignment.

4. Embedded Host Integration: Jetson, Raspberry Pi, ROS 2 & Edge AI Acceleration

Embedded Linux bring-up should establish what the camera actually delivers before adding ROS 2 or an inference runtime. On Jetson, Raspberry Pi or an industrial PC, start with the selected USB board or documented native receiver path, then record the camera revision, firmware, kernel, driver and power configuration.

For a UVC camera, the following read-only checks show the advertised formats and the current settings. Replace the device path with the verified camera node:

# Inspect the selected camera; these commands do not set a universal mode
v4l2-ctl --device=/dev/video0 --list-formats-ext
v4l2-ctl --device=/dev/video0 --get-fmt-video
v4l2-ctl --device=/dev/video0 --get-parm

Select a format and rate only from the supported combination. Do not force 640 × 512 Y16 at 50 Hz onto every Boson or CAMCUDA configuration. Confirm whether any telemetry rows are present and preserve the correct image dimensions, byte order and stride. Eight-bit AGC imagery may be appropriate for display or a model trained on that representation; quantitative analysis needs its documented data path.

In ROS 2, mono16 describes a single-channel 16-bit image representation. It does not declare that the pixel values are temperatures. Set encoding, endianness, row stride and image dimensions correctly. The Image message specifies an acquisition-time header timestamp; if only software receipt time is available, document that limitation rather than presenting it as measured acquisition time. Keep units, gain state and calibration metadata with the data path.

A conditional Boson R temperature example: FLIR’s T-linear output-conversion guidance gives Celsius = pixel value × 0.01 − 273.15 in high gain, and pixel value × 0.02 − 273.15 in low gain, for the described radiometric Boson T-linear output. First verify the radiometric part number, enabled T-linear mode and gain state. Do not apply these factors to ordinary pre-AGC counts, another core family or an unverified Boson+ configuration. Emissivity, window and atmosphere compensation, FFC, calibration conditions and measurement uncertainty still matter. Preserve the measurement samples separately from any display normalization.

Benchmark the actual model, preprocessing, accelerator runtime and thermal limits. Camera frame rate, unique image updates and inference throughput are separate measurements. Buffer sharing or DMA can reduce copies where supported by the driver/runtime; measure its effect on latency and sustained performance.

Log negotiated modes, sustained capture rate, sequence gaps, buffer age, host load, correction events and restart recovery. Repeat testing with the complete payload operating.

5. OEM Core Showcase: CAMCUDA AeroMini 640 & SuperMini Integration Paths

The CAMCUDA AeroMini 64060 Hz imaging is the primary interface-kit alternative in this guide. Its 640 × 512 VOx detector uses 12 µm pixels and publishes ≤30 mK NETD at 25 °C and F/1.0. Non-radiometric imaging is 60 Hz by default, with a 30 Hz factory option. The radiometric version is 25 Hz and enquiry-only; confirm its calibrated lens, data format and accuracy in the quotation.

AeroMini’s published 21 × 21 × 28 mm and <20 g figures exclude the lens and flange. Add the selected optics, tailboard, cables, mount and enclosure when comparing it with a Boson assembly. These figures do not describe a complete gimbal or a finished handheld imager.

Factory Lens Options & Optical Planning Reference

The following planning table uses current CAMCUDA product references; listed configurations are not a stock or delivery promise. The AeroMini imaging choices listed here are distinct from its radiometric enquiry options. SuperMini’s lens entries belong to a different 8 µm core; do not transfer their field of view or mechanical dimensions to AeroMini.

Use these published fields of view for initial comparison, then verify the selected lens, focus distance and required optical parameters in the design package. Geometric sampling alone cannot establish useful range or temperature accuracy. See the AeroMini lens table and SuperMini lens configurations for the corresponding product context.

AeroMini Boards, Cables & Software Resources

The USB + CVBS + MIPI tailboard and Type-C + CVBS tailboard are separate configurations. The illustrated 16-pin USB/CVBS and 26-pin MIPI/DVP references belong to the documented USB + CVBS + MIPI board. POWER_IN1 and POWER_IN2 are 5 V inputs. Do not connect them to 12 V or an unregulated vehicle power bus.

Read the schematic with the matching 16-pin signal table and the AeroMini datasheet, pages 3–4. The 26-pin drawing has its own MIPI/DVP assignments; it is not interchangeable with the 16-pin connector. Verify board orientation and serial-interface levels before wiring.

The AeroMini product FAQ points to a technical documentation folder containing manuals, USB client software and USB-SDK.zip, described by that FAQ as Linux drivers, examples and SDK resources. The archive has not been inspected or run for this guide. Match its contents to the ordered board, firmware, host and stream; this is not a claim of tested Linux or ROS 2 compatibility.

SuperMini 640 / 640T: Bare-Core Integration Contrast

CAMCUDA SuperMini 640 / 640TUltra-light core uses a 640 × 512, 8 µm detector. The 640 provides 50 Hz imaging; the 640T provides 30 Hz thermography. Its published ≤40 mK NETD is stated at 25 °C and F/1.0. The 13 × 13 × 13.4 mm and <3.5 g figures describe the bare core, excluding optics and boards.

SuperMini’s 30-pin core interface requires its own carrier design. MAIN_POWER at 3.8–5.2 V is only one supply: the documented +3.3 V and +1.8 V rails, tolerances, noise limits and power-on timing also apply. Core UART uses 1.8 V logic, with TX/RX defined from the core. CVBS requires an external video-buffer IC. The manual states that BT656 and MIPI cannot operate simultaneously; use the model-specific 640 BT656 or 640T CDS3/temperature-data definitions.

Review SuperMini power requirements and the selected expansion-board documentation before applying power. USB pins on the core do not establish a ready-to-connect USB kit. Request SuperMini-specific software resources rather than using the AeroMini SDK as compatibility evidence.

Review AeroMini 640 configuration options · Compare SuperMini 640 / 640T and request a configuration quote

6. Portable Field Instrument: Finished Handheld vs. OEM Core Integration

For an inspection team that needs a self-contained field instrument, a finished handheld thermal imager can avoid custom carrier, display and battery development. This is a different procurement choice from integrating Boson, AeroMini or SuperMini into a new product. The comparison below describes a generic finished-handheld category, not a current CAMCUDA handheld specification or an assembled version of these OEM cores.

Architecture & Procurement Checklist

A ready-made instrument can suit site walk-throughs and maintenance surveys. OEM integration serves custom robotic payloads, optics and data pipelines. Compare total cost and engineering effort at the same system boundary.

Define working distance, scene coverage, measurement needs, environment and data export. Confirm supplied items, power arrangements and lead time in the quotation; evaluate the complete instrument rather than its sensor alone.

7. Architectural Specification Matrix: FLIR Boson vs. CAMCUDA Solutions

Compare matched configurations: a Boson core, an AeroMini interface package, a SuperMini bare-core design and a finished handheld have different hardware boundaries and integration responsibilities.

Published NETD values describe sensitivity under stated conditions, not absolute temperature accuracy. The values above do not establish an independent head-to-head performance ranking. Price, availability, lead time and destination-specific supply requirements should be confirmed for the exact quotation.

8. SWaP-C Optimization, Optical Path Engineering & Johnson Criteria Sizing

SWaP-C planning should cover the complete inspection system: core, lens, interface board, cables, mount, enclosure, power conversion and host processor. A smaller bare core can reduce one part of that budget while requiring additional engineering elsewhere.

IFOV and scene sampling: A useful paraxial estimate is IFOV in mrad ≈ pixel pitch in µm ÷ focal length in mm. At a known distance, this gives an initial estimate of how much of the scene one pixel spans. Lens distortion, focus, optical resolution and viewing geometry can change the practical result; use the documented field of view and representative images to check the design.

The same focal length on 12 µm and 8 µm detectors produces different nominal sampling. Compare coverage, optical quality, noise and useful samples across the feature. Longer lenses also affect clearance, mass and mounting load.

Johnson criteria and range claims: A simple detection/recognition/identification model can be an early planning aid, but its assumptions are not a field-performance guarantee. Feature size, contrast, atmosphere, motion, optics, processing and the observer or algorithm all affect the outcome. The CAMCUDA thermal imaging calculator can support preliminary geometry comparisons; validate the intended civilian inspection task with the selected assembly. A predicted geometric range is not an accepted DRI specification for the ordered product.

F-number and sensitivity: A higher F-number generally reduces irradiance at the detector when the other optical factors are held constant. Transmission, spectral response, calibration, integration settings and processing also affect system sensitivity. Compare NETD only with its stated conditions, and do not scale a published core NETD into a guaranteed value for a different lens or assembled product. Temperature accuracy requires its own radiometric specification.

Heat flow and mounting: Follow the core’s mechanical and thermal guidance. Include nearby processors and power components in the thermal model; test warm-up, enclosure-temperature changes and airflow. Avoid core strain or shutter obstruction, and measure FFC/NUC behavior in the intended installation.

Use the SuperMini drawing above only for the bare-core scope. Its separate 6.1 mm lens-assembly drawing describes a different envelope. For AeroMini, obtain the dimensioned drawing for the exact lens, flange and board before releasing an enclosure; the linked datasheet supplies interface references rather than a universal dimensioned assembly drawing.

9. Deep-Dive Technical FAQ & Hardware Troubleshooting

What are the primary hardware challenges when interfacing thermal cores like FLIR Boson with Jetson or Raspberry Pi?

Start with connector mapping and logic levels, supply integrity, and the supported host format. A Boson core’s 3.3 V supply and 1.8 V EXT_SYNC are different electrical requirements. On Linux, enumerate the actual stream modes rather than assuming an 8-bit default or a universal Y16/rate combination. For AeroMini 64060 Hz imaging, choose the tailboard first and use its matched pinout; the illustrated POWER_IN1 / POWER_IN2 inputs are 5 V. Its published SDK resource does not establish tested compatibility with every Jetson, Raspberry Pi or ROS 2 stack. Validate sustained capture, controls and recovery on the chosen host.

How do developers configure multi-camera hardware synchronization (EXT_SYNC) for thermal stereo vision?

Use the exact core’s documented synchronization mode, signal direction and timing limits. The Boson bare-core EXT_SYNC in revision 380 is a 1.8 V signal; check any carrier translation before wiring. For supported fast R2.1-or-later configurations, section 6.17 describes rising-edge Slave operation and recommends a pulse width greater than 90 ns; follow the full timing specification and distinguish acquisition cadence from averaged output. Measure frame alignment and transport delay, account for FFC/NUC events, and document the timestamp meaning in the fusion software. Hardware sync does not by itself guarantee simultaneous scene response or sub-millisecond ROS 2 alignment, and its behavior must not be transferred to another core family.

What are the critical SWaP trade-offs when evaluating LWIR OEM camera cores for drone payloads and robotics?

Compare the entire assembly’s mass, envelope, electrical load and heat flow, including optics, interface boards and the host. AeroMini’s 21 × 21 × 28 mm and <20 g figures exclude lens and flange. SuperMini 640/640TUltra-light core‘s 13 × 13 × 13.4 mm and <3.5 g figures describe its bare core, excluding optics and boards. Its compact size comes with a 30-pin interface and multi-rail integration requirements. Measure the completed payload and its image behavior under the intended operating conditions; core specifications alone do not establish endurance, vibration performance or uninterrupted capture.

How does optical F-number affect LWIR sensor performance and NETD sensitivity?

Increasing F-number generally reduces irradiance when other optical factors remain equal. System NETD also depends on transmission, detector and processing conditions, so a simple aperture ratio cannot guarantee the final sensitivity. AeroMini publishes ≤30 mK and SuperMini ≤40 mK at 25 °C and F/1.0; these are configuration references, not temperature-accuracy claims or a tested ranking. Use lens-specific documentation and representative validation, especially if the task requires calibrated temperature measurement rather than visual contrast.