Long Waves, Short Talk

A practical IR spectrum introduction

Who is this guy?

• Spent the last 9 years working in the space industry as a systems engineer at NASA and a small satellite company.
• Photographer, which led to me getting into the fascination of seeing beyond the visible
• Enjoyer of restricted or otherwise limited hardware and information
• Owns too many thermal cameras (55 at last count)
• Prolific lowballer of weird and unusual scientific/government surplus

IR is perfect for this guy!

Call him hyperspectral the way he got 5 flavors of autism

What’s the deal with wavelengths?

• Will be discussing four wave bands in this talk scope
• NIR
• SWIR
• MWIR
• LWIR
• Visible will be left out because it’s the default experience, FIR is out of scope
• Infrared is a form of “light”, particle-wave duality and all that, and can be manipulated and detected as such

You are here

Talk ROI

Autism

Take your Tylenol now!

The Mk1 Mod0 Eyeball (unless you’ve had LASIK)

• You are using them to watch this talk right now!
• Neat pieces of hardware evolved over millennia
• Humans are limited to 700nm
• Perception is logscale, sensors are linear

Some spectral details

Idealized human emission spectrum

Baseline emission spectra to consider

Why are there notches in the sun spectrum?

• Common molecules (Water, Carbon Dioxide and Ozone) absorb specific wavelengths.
• The atmosphere has a LOT of those, other common gases also absorb
• But it gets even weirder, there’s also emissions
• Hydroxyl groups recombine after sunlight (UV) induced disassociation
• This recombination emits photons in SWIR

Water in SWIR, absorbing light

Skyglow hydroxyl emission spectra

NIR (and VIS) Band Imagery

Human dermis is partially transmissive in NIR

SWIR Band Imagery

MWIR Band Imagery

SXGA (1024*1280) 2016 InSb MWIR

LWIR Band Imagery

Okay we have a band we can’t see, how do we detect it

• Four major types of detectors
• Thermopile
• Poly-Si/Aluminum
• Microbolometer
• Vanadium Oxide
• Photovoltaic/photoconductive
• III-V elements
• Superlattice (out of scope of talk)
• III-V elements, but way wilder processes and Quantum nightmare
• IR is mostly heavy metals and toxic materials

Vocabulary to remember: NETD (Noise Equivalent Temperature Difference) is the smallest temperature difference a thermal camera can reliably detect — in other words, it tells you how sensitive the camera is to tiny changes in heat.

​

Performance

COST

SCARY DETECTIVITY GRAPH JUMPSCARE�

Silicon Visible-NIR sensors.

• The classic system everyone here is familiar with
• Silicon photodiodes and CMOS readout with amplifiers in pixels
• Hard stop of usability around 1.2 microns, as the silicon becomes transparent
• Some more interesting designs can hit spec of 40% qE at 1 micron
• This is remarkable for a cheap sensor, but still limited in a variety of ways
• Stated sensor is the Sony IMX462, decently available as USB modules and Raspberry pi compatible modules but NDA-locked full sheets.
• Beyer filter dyes can be tuned to be transparent at NIR, thus not impeding the band photons
• Deeper Quantum well depth

Thermopile

• Simplest and cheapest method of detecting thermal radiation
• Fabricated in standard one-shot MEMS single wafer process
• Slow time constant (order of high 10s of mS)
• High baseline NetD linked to read out speed
• Heimann specifies 50-200mK @1hz
• 𝑁𝐸𝑇𝐷@1 𝐻𝑧 × sqrt(𝐹𝑟𝑎𝑚𝑒𝑟𝑎𝑡e)
• Degrades the faster you go
• Large pixels, large pixel pitch
• Limits maximum resolution
• Highest seen yet is 160*120
• Optics design becomes harder
• Best used for PIR sensors
• Available in MLX90614

​

Diagram of a detector element

Actual Detector element (Heimann)

Microbolometer

• Most widely used detector technology in cost effective COTS systems.
• Requires a ROIC(Read Out IC) and hybridization (Indium bump bonding)
• Moderate time constant (7-12mS)
• VOx(Vanadium Oxide) and ASi(Amorphous Silicon) are two most common materials
• ASi sees some degradation over longer term due to Staebler-Wronski effect (out of scope)
• Improved NetD as technology matured
• 30-50mK is typical range
• Pixels shrunk, pitch 8-20um
• Requires routine NUC (non-uniformity correction)
• Resolution 80*60 – 1280*1024
• Cost range $120-$30K
• Must be vacuum encapsulated
• Vacuum keeps response faster, prevents oxidization/moisture
• This packaging is a major cost driver
• WLP has highest NRE

Diagram

Actual

Direct conversion (PV/PC) III-V

• High performance, expensive and fast as all hell
• Requires a ROIC(Read Out IC) and hybridization (Indium bump bonding detector to ROIC)
• Short time constant (700ns or less)
• InGaAs, InSb, PtSi, HgCdTe are common III-V combinations.
• Best in class NetD of around 10mK
• Pixel pitch around 8-20um
• No NUC required except during production or at lens swap.
• Resolution 320x240 – 2048*2048
• Cost range $15K-$500K
• Requires cooling to sit in the bandgap.
• Cryogenic temperatures of 70K or for HOT 140K.
• Sterling cryocooler engines are large cost drivers
• IDCA(integrated dewar cooler assemblies) $$$
• Cold shields are a must to reduce environmental noise
• Hard vacuum and Helium hell
• If you’ve never worked with Helium, it is leak city

Detector and ROIC

Cold Shield

Bondwires to fanout

Looking into the dewar of a 1996 DOM early 256*256 InSb IDCA

Colormap: WHOT (Black cold)

SCARY Quantum Efficiency Graph JUMPSCARE�

What Even Is a ROIC (the oversimplified version)

• Silicon chip that has biases, clock and amplifiers to drive and return pixel values, connected to detector array via Indium bump bond fields.
• If it is a DROIC it will have the ADC as well, most modern will have this style.
• Either CMOS style below-pixel TIA or CCD style multiplexed amplifier.
• 90-130nm tech node most common
• Indium bumps to connect to other detector substrates.
• I’d include images of my AIM-9X ROIC but those are probably controlled still

Yeah that’s detectors, but what about optics

• For Vis-NIR (and sorta SWIR) you can use standard BK7 and other glasses.
• SWIR really prefers Fused Silica or Sapphire
• Issues start around 1.1 microns, where most lenses are not designed to converge and focus, and glasses stop working well
• For thermal bands, you MUST use exotics such as metals (Si, Ge, GaAs), chalcogenides or Metal combos (ZnSe, BaF2, MgF2)
• These exotics drive the cost to insane levels due to cost of material (Ge is around $3-10 a gram and MUST be single point diamond turned in an optical lathe). Chalcogenides are moldable allowing for lower relative cost.
• Pictured zoom lens for LWIR was purchased for $1K, MSRP of $55K for a low-end zoom.
• Other specialist optics (microscope, wide angle etc) cost lots of money.
• To improve circuit debugging use and approximate a microscope lens you can buy a cheap $20ish ZnSe lens and slap it in front of a normal LWIR camera to improve up close performance.

$35K SWIR Zoom

$55K LWIR Zoom

See here

What does this all mean for the home gamer?

• Thermal cameras have cratered in price and are incredibly accessible to the hobbyist.
• At $500 or under you can get fantastic performance from US or Chinese detectors off of Amazon/Ebay/Aliexpress (up to 320*240@25hz)
• Under $250, the whole market is Chinese, specifically the Tiny1C (256*192@25hz) from Raytron(used to be IRay, and several other shells to get around sanctions, due to the Russo-Ukranian war)
• Good news is there is extensive python/opencv support for most tiny1 based cameras, they enumerate as UVC webcams.
• Beyond this, you are going to be in an uphill battle as most industrial or otherwise need extensive technical know-how or are just reverse engineering targets.
• Do not recommend surplus cameras, more often they are just trouble, requiring expensive software or are black boxes of unknown communications
• MWIR is a special hell, as most companies will not discuss them without being a company that is ITAR registered and EUC in place, as these are intensely dual-use items.

Tiny1C based P2 Pro image

Questions?

Find me at DiscountMissiles.com!!!