Weekly Patent Review

TSMC patent application US20080014673 talks about a way to improve crosstalk between pixels. Even though the idea is applicable to both backside and frontside illuminated sensors, most of the figures relate to the backside version.

The isolation between pixels is achieved by aluminum doped deep walls. A thermal process is employed to drive aluminum deep into the substrate. The substrate has a <100> or <111> orientation such that aluminum diffusion along the direction perpendicular to the substrate is much higher than aluminum diffusion to other lateral directions. Therefore, the aluminum doped deep wall can be formed without much lateral distortion.

Height of the aluminum doped deep wall is determined by thickness of the substrate. Preferably the height of the aluminum doped deep wall is greater than one-fourth of the thickness of the substrate to reduce crosstalk. For example, thickness of the substrate is about 4 micron, height of the aluminum doped deep wall will be greater than about 1 micron. The aluminum doped deep wall may have a ratio of depth/width (or height/width) greater than about 3.

The thermal process of driving aluminum may be a thermal annealing process performed by a tool such as a rapid thermal process (RTP) tool or a flashing annealing tool or a laser beam. The annealing temperature may range between about 400C and 600C.

If this works, it might significantly improve the quality of the backside illuminated pixels. As a sidenote, this application reveals that TSMC spends its resources on backside illumination process development.

Samsung application US20080012973 enhances dynamic range of the sensor by making 4T photodiode with two transfer gates, one - for low light level, the other - for high illumination. The main idea is to increase the charge storage capability by integrating the signal on a bigger floating diffusion capacitance for large signal, while the small signal gets advantage of higher conversion factor. I've seen many similar ideas in the past. Samsung's idea is a little different in details, but in any case I see it as of very limited use.

Canon's application US20080012965 talks about accurately subtracting dark current. To reduce infuluence of hot pixels and other artifacts, Canon algorithm virtually builds a histogram of the dark signal distribution and tries to eliminate pixel outliers. It should work, albeit many similar algorithms work too.

Micron's application US20080012966 improves row noise reduction circuit. Dark columns are usualy used to suppress supply and substrate noise by providing a reference signal for subtraction. If dark columns contain hot pixels or bad pixels, the reference signal acquires a row fixed pattern noise. To prevent that Micron proposes to use a separate dark section in the array, rather than full columns. Having separate TX, SEL and RST controls, this section provides a row address independent reference, so row FPN is eliminated. Sounds like a valid idea to me, with one exception. The separate TX, SEL and RST pixel controls have different length, thus row driver load and timing are different. This might affect the sampled supply noise, so row noise would re-appear.

Another Micron's application US20080012971 uses the old analog trick to choose a quite sampling time when the system supply and substrate noise is minimal. The only twist I'm able to spot is usage of a programmable microcontroller or logic to set the right clock delay for sampling. To me this is not a big deal.

Visera's application US20080011936 proposes to change microlens curvature across the array. I'm not sure the idea itself is new, but if Visera can really manufacture such microlenses, it opens a whole lot of interesting opportunities.