Three Thousand Frames a Second at One Megapixel

Teledyne e2v has extended its Flash CMOS global-shutter image sensor family with two models. The Flash 1K2, at 1K by 1K, captures up to 3,000 frames per second at full frame, aimed at rapid event capture and high-speed measurement. The Flash 8K, at 8K by 1K, runs at 1,800 fps for 3D laser profiling and high-resolution inspection. They join the existing Flash 2K (2K by 1K at 1,500 fps) and Flash 4K (4K by 1K at 1,800 fps), giving a family spanning 1K to 8K on one platform with a common architecture and development tools. All use a 6 micrometre global shutter pixel with high dynamic range to 100 dB, programmable regions of interest, frame-to-frame adjustability and a single-tap design.

The 8K by 1K geometry is the one that explains the product. That is not a general-purpose sensor shape; it is a laser triangulation sensor shape. In 3D profiling, a laser line is projected across an object and imaged at an angle, and the sensor only needs to be tall enough to capture the line's displacement while being as wide as the object and as fast as the conveyor. Resolution across the line sets lateral resolution; frame rate sets how finely the object is sampled along its travel. At 1,800 fps a part moving at one metre per second is profiled every 0.55 millimetres, which is what makes the number meaningful.

Global shutter is the non-negotiable part of that and is worth restating because rolling shutter sensors are cheaper and constantly proposed. A rolling shutter exposes rows sequentially, so anything moving is captured at slightly different times down the frame, which skews the geometry — fatal when the geometry is the measurement. Every moving-object metrology application needs global shutter, and the 6 micrometre pixel here is a deliberate compromise: large enough to collect useful signal at short exposures, small enough to keep the 8K die economic.

The 100 dB dynamic range is the specification that quietly determines whether an inspection works on real parts. Industrial scenes routinely contain a specular highlight from a machined surface and a shadowed recess in the same frame, and a sensor that clips one or loses the other produces a measurement gap exactly where the feature of interest sits. For anyone evaluating these, the useful test is not a resolution chart but the worst part in the production mix — the shiniest, the darkest, and the one with the deepest bore — because that is the part that will define the line's false reject rate.