Friday, June 18, 2010

Omnivision Proposes Local Laser Annealing in BSI

The recently published Omnivision patent application US20100140675 discloses local laser annealing idea on the sensor backside:

Here are the advantages of this proposal:

"Reducing the silicon thickness introduces process complexities in terms of laser annealing the backside dopant layer. ...the temperature rise at the front surface will be greater for thin silicon. Accordingly, in some embodiments, a laser anneal mask (not illustrated) is deposited on the die or wafer backside and patterned to only expose those areas immediately below PD regions 415. When the annealing laser is scanned across the die or wafer backside (see FIG. 6C) only the uncovered portion of dopant layer 405 is exposed to the laser, so only the boron in this exposed area is activated. The thickness and composition of the laser anneal mask are laser and process dependent; however, the composition and thickness should be chosen to have a high reflectively to the laser wavelength.

Masking the die backside prior to the laser anneal of dopant layer provides a number of advantages. The amount of energy absorbed by the silicon is reduced, thereby reducing the temperature rise of the die. There are typically no metal/silicide contacts above photodiode region 415 and therefore the front surface can tolerate a larger temperature rise than regions with metal contacts. Metal contacts will typically degrade above 400 C, but dopant profiles typically do not diffuse below 800 C. The periphery circuit around photodiode region 415 and pixel array is protected from the harmful effects of the laser anneal process. The P+ type ion implant under the masked back surface area is not exposed and therefore is not activated creating a high recombination region. Electrons that migrate to or are formed in this region will recombine easily reducing crosstalk.

There is also "a backside surface treatment layer is applied to further passivate the backside of epi layer and cure surface defects... Surface treatment may include exposing the backside silicon surface to a gas or liquid containing a reducing or oxidizing species to reduce backside surface states. These species may include H2, H2O, N2O, O2, O3, and H2O2, among others. Curing defects and reducing surface states advantageously can reduce dark current and white pixels."


  1. Albert TheuwissenJune 18, 2010 at 8:50 PM

    Strange solution : part of the back-side will be passivated, part of it not. The unpassivated part will still generate dark current and will still have a low QE (in blue). So what are the benefits (besides avoiding stitching issues of the laser anneal) ?

  2. Initially I interpreted the patent that the photodiode area is in fact the whole pixel array area. So, the pixel array is passivated while periphery is not. Now, as I re-read the description, I'm not sure that my interpretation was correct. I agree with you that placing the passivation mask over the pixel array is a mistake.

  3. @ placing the passivation mask over the pixel array is a mistake.

    I meant dividing the array into passivated and un-passivated regions is not a good idea.

  4. The application does seem to indicate passivation. The posted figure looks like a way of protecting the front-side vias from getting too much heat when a back-side boron implant is annealed on very thin silicon(they discuss 1.5 um to 3 um thicknesses). Passivation is mentioned as a later processing step.

    Most of Chicagoland just got annealed by a massive thunderstorm, and on the backside of it I'm going to go implant the kids in the local pool during free family swim hour.

    Go USA! (My obligatory World Cup plug.)

  5. My problem is with this phrase:

    "The P+ type ion implant under the masked back surface area is not exposed and therefore is not activated creating a high recombination region. Electrons that migrate to or are formed in this region will recombine easily reducing crosstalk."

    The problem is that this type of recombination inevitably goes along with generation. So, if inter-pixel regions are left with no annealing, the dark current would be higher.

  6. Albert TheuwissenJune 19, 2010 at 4:53 PM

    Right, and QE will be lower.

  7. If the high recombination interfaces have no interaction with the depletion regions,which seems to be the case, dark-current, IMHO, will not be an issue (at normal working temperatures). The QE will suffer in the shorter wavelengths, but not that much for longer, while the cross-talk will be reduced (blue-photons create worst CT in BSI). Moreover, parasitic light on the floating node/ckts seem also to be reduced.

  8. I think that this is for reduce the pixel to pixel crosstalk. In the old FPA, we used proton implant to create pixel isolation. A high energy implant creates also amorphe surface. The fact that the local laser annualing will only make re-crystalization on the photodiode region and create delta doping profile. The non activated region will have a amorphe surface and have NO Delta doping profile, so the surface generated dark current will have small chance to migrate to the photodiode region.

    The lower QE can be recovered by microlens. For example, the metal shield on SONY sensor reduces also the QE, so microlens is always necessary. Here the situation is totally different from scientific backthinned sensors ...


  9. This comment has been removed by the author.

  10. I think you guys are neglecting the possibility that sometimes patents are filed for things that don't quite work well, or may have a lot of claims but only one or two gems. When you sit on a patent committee, for example, when it is unclear if something might be useful or important, you still prefer to err on the side of applying for a patent.
    This still needs to go thru examination and then perhaps if it is useful and if someone else uses it, it needs to be defensible in court as being either not obvious and not covered by 40 years of backside-illuminated image sensor history.
    There are a lot of silly or poorly written, or poorly examined patents out there.

  11. well we are justing analysing the intention of this patent application. Of course the only patents possible on BSI will be on the details of processing. The basic principle is of course well known. 99% of patents are poorly written or of no real invention inside. I fully agree with Eric

  12. Actually, come to think of it, the discussion above by knowledgeable engineers, expert even, pretty much makes the case that the invention was NOT obvious.

  13. Ironically, the weak or marginal patents are quite valuable in large numbers, as they deter infringement lawsuits with the threat of counter-suits. Negotiating a cross-license is usually much cheaper than going through a full trial.

    For the little guy, this makes it hard to get a new patented idea tried out.

    (By the way, thanks to the several senior forum participants who've been willing to take a look at my little-guy work.)


All comments are moderated to avoid spam.