Press release: https://www.sony-semicon.com/en/news/2025/2025061001.html
Sony Semiconductor Solutions to Release Stacked SPAD Depth Sensor for Automotive LiDAR Applications, Delivering High-Resolution, High-Speed Performance High-resolution, high-speed distance measuring performance contributes to safer, more reliable future mobility
Atsugi, Japan — Sony Semiconductor Solutions Corporation (SSS) today announced the upcoming release of the IMX479 stacked, direct Time of Flight (dToF) SPAD depth sensor for automotive LiDAR systems, delivering both high-resolution and high-speed performance.
The new sensor product employs a dToF pixel unit composed of 3×3 (horizontal × vertical) SPAD pixels as a minimum element to enhance measurement accuracy using a line scan methodology. In addition, SSSs proprietary device structure enables a frame rate of up to 20 fps, which is the fastest for such a high-resolution SPAD depth sensor having 520 dToF pixels.
The new product enables the high-resolution and high-speed distance measuring performance demanded for an automotive LiDAR required in advanced driver assistance systems (ADAS) and automated driving (AD), contributing to safer and more reliable future mobility.
LiDAR technology is crucial for the high-precision detection and recognition of road conditions and the position and shape of the objects, such as vehicles, pedestrians. There is a growing demand for further technical advancements and developments progress in LiDAR toward Level 3 automated driving, which allows for autonomous control. SPAD depth sensors use the dToF measurement method, one of the LiDAR ranging methods, that measures the distance to an object by detecting the time of flight (time difference) of light emitted from a source until it returns to the sensor after being reflected by the object.
The new sensor harnesses SSS’s proprietary technologies acquired in the development of CMOS image sensors, including the back-side illuminated, stacked structure and Cu-Cu (copper-copper) connections. By integrating the newly developed distance measurement circuits and dToF pixels on a single chip, the new product has achieved a high-speed frame rate of up to 20 fps while delivering a high resolution of 520 dToF pixels with a small pixel size of 10 μm square.
Main Features
■ Up to 20 fps frame rate, the fastest for a 520 dToF pixel SPAD depth sensor
This product consists of a pixel chip (top) with back-illuminated dToF pixels and a logic chip equipped with newly developed distance measurement circuits (bottom) using a Cu-Cu connection on a single chip. This design enables a small pixel size of 10 μm square, achieving high resolution of 520 dToF pixels. The new distance measurement circuits handle multiple processes in parallel for even better high-speed processing.
These technologies achieve a frame rate of up to 20 fps, the fastest for a 520 dToF pixel SPAD depth sensor. They also deliver capabilities equivalent to 0.05 degrees vertical angular resolution, improving the vertical detection accuracy by 2.7 times that of conventional products. These elements allow detection of three-dimensional objects that are vital to automotive LiDAR, including objects as high as 25 cm (such as a tire or other objects in the road) at a distance of 250 m.
■ Excellent distance resolution of 5 cm intervals
The proprietary circuits SSS developed to enhance the distance resolution of this product individually processes each SPAD pixel data and calculates the distance. Doing so successfully improved the LiDAR distance resolution to 5 cm intervals.
■ High, 37% photon detection efficiency enabling detection of objects up to a distance of 300 m
This product features an uneven texture on both the incident plane and the bottom of the pixels, along with an optimized on-chip lens shape. Incident light is diffracted to enhance the absorption rate to achieve a high, 37% photon detection efficiency for the 940 nm wavelength, which is commonly used on automotive LiDAR laser light sources. It allows the system to detect and recognize objects with high precision up to 300 m away even in bright light conditions where the background light is at 100,000 lux or higher.
Well done!
ReplyDeleteThat's a strange article. What's with the repetition of those 520 dTof pixels? It has 105x1568 pixel which are binned to 35x521 (HxV) = 18,235 pixel in total.
ReplyDelete520 dToF pixel --> 35x521. The 521 is the "520 dToF" pixels.
DeleteWhy are they highlighting the vertical # of pixels instead of both horizontal and vertical? I am assuming it is because of the readout scheme that might be column-based and compatible with horizontal scanning. So according to your laser source, you can read a total 1x520 dToF pixels at a time and scan to cover Mx520 pixels total.
What concerns me is the 300m ranging claim. This is a receiver for a LiDAR module. It does not contain a transmitter. So, claiming that his enables 300m ranging is misleading. Regardless of the good PDE, you need laser power to detect anything. Laser power, laser pulse shape, laser scanning all impact the ranging performance. So, I would be curious to see in which conditions this can actually reach 300m.
35 pixels are used to recover laser line alignment error. If the laser is perfectly aligned, then 1x520 are enough.
DeleteOne question: how the grayscale image is formed? By photon counting with the lidar laser source?
DeleteProbably not. The grayscale image seems from a different camera with different field of view.
DeleteThe grayscale image would likely be formed by counting the interference generated by existing light in the scene i.e. effectively noise from the sun. You can tell as the buildings in the background and the sky have a textured gradient. The intensity information from the laser source is shown as the dark blue to yellow gradient with red being retro reflectors. It comes from the same sensor the FOV is the same just the LiDAR view has been reprojected.
Delete@AnonymousJune 22, 2025 at 2:17 PM
ReplyDeleteA SPAD sensor builds a histogram with the received data. 300m max distance means the histogram is big enough for 300m (but not more). Every one of the 520 pixel has its own histogram. This takes a lot of RAM, so they won't make it unnecessary big.
A common requirement/wish from the car manufacturers is they want to see a car tire lying on the road 200m away. If you can do this, then 300m on a reflector (number plate) can't be too difficult.
make sense!
DeleteThat's true, but the strong signal from a reflector can cause aliasing and the return can appear (falsely) at much shorter distance than 300m causing false alarms if not carefully paired with an intensity image.
DeleteThe easiest thing to do is to build the gray scale image from the noise floor. Because you could potentially build that regardless of your illumination and you wouldn't need to adjust for the laser return intensity.
DeleteHow the grayscale image is generated? is it done by using the laser illumination or just ambient light?
ReplyDeleteAmbient light. Ouster all do this
Delete