Picture sensor embeds AI to discover picture knowledge

A brand new technology of CMOS picture sensors can exploit all of the picture knowledge to understand a scene, perceive the state of affairs, and intervene by embedding synthetic intelligence (AI) within the sensor. CEA-Leti researchers have reported this design breakthrough when demand for sensible picture sensors is rising quickly resulting from their high-performance imaging capabilities in smartphones, vehicles, and medical gadgets.

The design breakthrough is constructed on a mixture of hybrid bonding and high-density by way of silicon by way of (HD TSV) applied sciences, which facilitates the mixing of assorted elements like picture sensor arrays, sign processing circuits and reminiscence components in picture sensors with precision and compactness.

The design breakthrough is predicated on a three-layer check car that featured two embedded Cu-Cu hybrid-bonding interfaces, face-to-face (F2F) and face-to-back (F2B), and with one wafer containing high-density TSVs. Supply: CEA-Leti

05.31.2024

05.30.2024

05.28.2024

Communication between the completely different tiers in a picture sensor design necessitates superior interconnection expertise. The brand new design offered by CEA-Leti employs hybrid bonding resulting from its very nice pitch within the micron and sub-micron vary. It additionally makes use of HD TSV, which has the same density that permits sign transmission by way of the center tiers.

“Using hybrid bonding and HD TSV applied sciences contribute to the discount of wire size, a crucial think about enhancing the efficiency of 3D-stacked architectures,” stated Renan Bouis, lead writer of the paper titled “Bottom Thinning Course of Improvement for Excessive-Density TSV in a 3-Layer Integration.” He added that stacking a number of dies to create 3D architectures, similar to three-layer imagers, has led to a paradigm shift in sensor design.

The paper presents the important thing technological bricks which can be necessary for manufacturing 3D, multilayer sensible imagers able to addressing new purposes that require embedded AI. “This units the stage to work on demonstrating a totally purposeful three-layer, sensible CMOS picture sensor, with edge AI able to addressing high-performance semantic segmentation and object-detection purposes,” stated Eric Ollier, mission supervisor at CEA-Leti and director of IRT Nanoelec’s Good Imager program.

The Grenoble, France-based analysis home CEA-Leti is a significant accomplice of IRT Nanoelec, an R&D institute additionally based mostly in Grenoble, France.

It’s value mentioning that at ECTC 2023, CEA-Leti scientists reported a two-layer check car combining a 10-μm excessive, 1-μm diameter HD TSV and extremely managed hybrid bonding expertise, each assembled in F2B configuration. Now, they’ve shortened the HD TSV to 6-μm top, which led to the event of a two-layer check car exhibiting low dispersion electrical performances and enabling easier manufacturing.

It’s primarily resulting from an optimized thinning course of that allowed the substrate thickness to be lowered with favorable uniformity. “This lowered top led to a 40% lower in electrical resistance, in proportion with the size discount,” stated Stéphan Borel, lead writer of the paper titled “Low Resistance and Excessive Isolation HD TSV for 3-Layer CMOS Picture Sensors”. “Simultaneous decreasing of the facet ratio elevated the step protection of the isolation liner, resulting in a greater voltage face up to.”

Scientists at CEA-Leti are assured that this sensible picture sensor expertise will allow a wide range of new purposes.

Associated Content material