On September 10, 2026, Singular Photonics, based in Scotland, UK, announced "Litavis," an image sensor using single-photon avalanche diodes (SPADs). It continuously counts photons across a 256×256 pixel array while also recording photon arrival times in 64×64 units. The company claims a "world first" for integrating imaging, timing, histogramming, and photon statistics through in-pixel processing on a single chip. However, SPADs and in-pixel time-to-digital converters (TDCs) themselves are not new. To judge what is actually new here, we need to separately examine what the two pixel arrays can output simultaneously.
Bundling 65,536 Detectors into 4,096 Timing Units
Litavis's photosensitive surface contains 256×256 SPADs arranged at a 10.17µm pitch. Each SPAD operates with a reverse bias exceeding its breakdown voltage, amplifying the charge generated by a single photon in an avalanche-like process and converting it into a digital pulse. The number of pulses counted over a fixed time represents light intensity, while the timing of pulse arrivals can be used to estimate distance or fluorescence lifetime.
On the timing side, groups of 4×4 adjacent SPADs are combined into a single macropixel. The 256×256 SPAD array totals 65,536 detectors, while the 64×64 macropixel array totals 4,096 units—meaning each macropixel bundles 16 SPADs, a figure derivable from the array dimensions given in the datasheet. As a result, intensity images are obtained at 256×256 resolution, but depth or lifetime maps built from arrival times are limited to a 64×64 grid. Timing information is not output at the same resolution as intensity.
Still, there is practical value in being able to output continuous photon counting from 65,536 SPADs and low-rate, time-stamped events from 4,096 macropixels in parallel. This is because both intensity and timing can be extracted from the same photosensitive surface in situations that would previously have required switching detectors or readout systems depending on measurement conditions. The 3.5mm-square sensor is a 3D-stacked, backside-illuminated device combining 65nm and 40nm CMOS processes, with the photosensitive layer separated from the processing layer. This design eases the problem of the photosensitive area and digital circuitry competing for the same plane, but the process details, yield, and manufacturing cost for each layer have not been disclosed.
37 Picoseconds Is Not the Whole Camera's Accuracy
The datasheet's figure of approximately 37ps refers to the width of a single tick mark by which the TDC divides time. Since one picosecond is one trillionth of a second, this may look like a very fine number, but it cannot be directly translated into LiDAR distance accuracy or fluorescence lifetime measurement precision. The overall response—including the light source's pulse width, detector jitter, and readout circuitry—needs to be verified using a different metric.
37ps is the TDC's time bin width, not the sensor's overall response width. The datasheet lists the average IRF (full width at half maximum) as 224ps at 640nm and 1145ps at 450nm. IRF represents the spread of the observed time response when a short light pulse is input. However, the datasheet does not disclose details of the measurement light source, bias, temperature, or statistical processing, so the difference between these two listed values cannot be attributed to wavelength alone. In other words, the 37ps increment demonstrates the ability to record time finely, but it does not guarantee that individual photon arrivals can be reproduced with 37ps resolution.
This distinction also shows up in prior research. A 192×128 pixel SPAD sensor presented in 2019 in the IEEE Journal of Solid-State Circuits by Singular Photonics co-founder Robert Henderson and colleagues operated an in-pixel 12-bit TDC with steps of 33 to 120ps, yet had an average IRF of 219ps. In-pixel TDCs and tens-of-picosecond time steps are not technologies appearing here for the first time. Litavis's "world first" claim should be read as the company's qualified assertion of combining multiple measurement modes simultaneously on a single chip.
Even Single-Photon Sensitivity Has an Appropriate Brightness Range
Litavis's photon detection efficiency (PDE) is stated as 47% under conditions of 785nm wavelength, 21V substrate bias, and 20°C. PDE is the proportion of incident photons that become detected pulses, and it does not mean 47% across all wavelengths or all operating conditions. Similarly, the 100% fill factor with microlenses is a geometric specification describing how light on the pixel surface is directed to the effective photosensitive area. It does not indicate that 100% of incoming photons are detected.
Once a SPAD fires, it cannot receive the next photon until the avalanche is quenched and the voltage is reapplied. Litavis's dead time is 4.3ns. In scenes that are too dark, an insufficient number of photons lowers the signal-to-noise ratio; in scenes that are too bright, photons arriving during the dead period go uncounted. Time histograms can also suffer from "pile-up," where earlier-arriving photons are recorded disproportionately often. Being able to detect down to a single photon is not the same as being able to measure linearly from darkness to bright light.
The dark count rate (DCR), which occurs even without light, is listed as approximately 60 counts/s at 18.8V and 20°C. However, it cannot be determined from the datasheet whether this is a mean or median value, or whether it is measured per SPAD or in some other aggregation unit. This makes it unusable for ranking performance against other sensors measured under different conditions. Crosstalk, afterpulsing, and pixel-to-pixel variation are also undisclosed, leaving open questions for applications that integrate weak light over long periods.
Software Design That Turns One Device from LiDAR into a Microscope
Litavis switches between photon counting, HDR, time-correlated single-photon counting (TCSPC), multi-event histogramming, and time gating via software. TCSPC works by repeatedly emitting pulsed light and accumulating photon arrival times into a histogram. For LiDAR, this converts round-trip time into distance; for fluorescence lifetime imaging (FLIM), it converts the decay of emitted light into a lifetime value. Time gating selects only a specific time window, suppressing fluorescence backgrounds that arrive with delay, as in time-resolved Raman spectroscopy.
TCSPC's measurement range is approximately 1µs at 15 bits, while multi-event histogramming has 8 or 16 bins per pixel. The latter's bin width is listed separately as ranging from 300ps to 614.4ns, with a maximum Timing Range of 9.8ms. Because the datasheet does not explain the relationship between bin count, bin width, and maximum range, the 9.8ms figure cannot be treated as simply the product of 16 bins and 614.4ns. Standard photon counting operates at 256×256 pixels with 14 bits, while HDR mode operates at 128×256 pixels with 28 bits—halving horizontal resolution to extend the counting range.
Building counts and histograms within the pixel eliminates the need to stream every photon event off-chip, potentially reducing data volume. In addition to a GUI, the company provides a general-purpose API through a C-language foreign function interface and a Python SDK. The apparent aim is to make it easier for researchers to reconfigure a single unit for multiple experiments. However, comparative figures for external data transfer volume, processing latency, maximum frame rate, maximum event rate, and power consumption have not been published, so the claimed reduction in bandwidth or processing load cannot yet be verified.
Even After the Product Announcement, the Numbers Needed for Adoption Are Missing
The finished camera measures 102×145×85.25mm and includes a 12V power supply, USB-C connectivity, and four factory-configurable SMA sync ports. Singular Photonics states it has received a "substantial volume" of pre-orders from companies and research institutions, but it has not disclosed unit numbers, customer names, pricing, or a general shipping date. Announcing a product should be treated separately from the state of anyone being able to estimate delivery time and purchase it.
There are signs of commercial progress. The company, a spinout from the University of Edinburgh, raised $2.15 million in August 2026. It has also stated that its year-to-date revenue in 2026 has reached twice the total for all of 2025. However, no absolute figures are given, and it's unclear what share of revenue Litavis accounts for. While pre-orders and revenue growth indicate the existence of customer engagement, they are not figures that prove mass-production capacity or profitability.
Whether Litavis can become a common platform replacing dedicated sensors is not determined by how many application names it can list. For LiDAR, distance error and ranging range must be shown; for FLIM, the reproducibility of lifetime estimates; and for spectroscopy, the signal-to-noise ratio after background removal—all alongside illuminance, wavelength, and integration time. Only once pricing, shipping timelines, speed, power consumption, temperature range, and crosstalk are all disclosed, and third parties can reproduce results for each application, will it be possible to weigh the value of switching functions via software against the cost of adoption.