Monday, February 08, 2010

Omnivision Announces 1.1um BSI Pixel

OmniVision announced the introduction of the world's first 1.1-um BSI pixel, OmniBSI-2. Proclaimed OmniVision's second-generation BSI technology, the is pixel is built on a 300-mm copper process at 65 nm design rules developed in cooperation with TSMC.

An exact quote from the PR: "By combining custom 65 nm design rules and new manufacturing process modules, the 1.1-micron OmniBSI-2 pixel achieves industry-leading low-light sensitivity as well as significantly reduced dark current and full-well capacity."

[Update: The newer versions of the PR fixed the typo and now talk about "improved", rather than "reduced" full well.]

OmniBSI-2 pixel design features layout, better isolation, and significantly reduced crosstalk over the first generation OmniBSI technology. OmniBSI-2 technology can also be applied to larger pixel designs to achieve performance advantages over current BSI and FSI image sensors.

"By comparison, the new 1.1-micron OmniBSI-2 pixel not only outperforms our current 1.75-micron FSI architecture, but it also equals the performance of our industry-leading 1.4-micron BSI pixel that is currently in mass production," commented Dr. Howard Rhodes, vice president of process engineering at OmniVision. "Migrating to 1.1-micron BSI pixel architecture required moving production to TSMC's state-of-the-art 300-mm copper process, which enabled substantially improved design rules and more advanced process tools, resulting in tighter process control and improved defect density. Key to our success was the joint development by the OmniVision and TSMC R&D teams of multiple new process modules that substantially improved opto-electronic performance. We also leveraged our close partnership with joint venture partner VisEra Technologies to establish a 300-mm color filter fabrication capability."

OmniVision's OmniBSI-2 technology will be demonstrated to customers by appointment at the Mobile World Congress, February 15 through 18, 2010 in Barcelona, Spain. No word is said about products to be demonstrated or their status (prototype, CS, ES, production, etc).

As Omnivision web site shows, the main difference between OminBSI and OmniBSI-2 are increased light sensitivity, better full well and higher QE:


  1. I remember someone from OV telling me that 12" wafer starts were too expensive, and that was why they liked BSI so much. It enabled them to stay on 8" wafers. Now they have moved to 12". I wonder if TSMC gave them a deal on 12" or they just couldn't shrink the pixel anymore without going to copper.

    Also, I'm not sure what Howard means by equals the performance of our 1.4 BSI. Is the original BSI all that great? I've heard that it doesn't do low light worth a darn.

    FWIW, I'm not down on OV, but I've learned to take these press releases with a big grain of salt.

  2. If you started about grains of salt... Howard says 1.1um BSI-2 outperforming 1.75um FSI. I recall the times when Omnivisions said that 1.75um pixel is substantially better than 2.2um, here for example:

    Combining these two statements together, 1.1um BSI should outperform 2.2um FSI too.

  3. The PR says "improved" dark current and full well capacity, not "reduced"...which would not be something to brag about.

    Meanwhile, I suppose when HR means "outperforms" he means QE, full-well density, and perhaps overall dark current, and probably not SNR10, absolute FWC, or dark current density. But who knows, someone should just ask him directly.

  4. What do 300mm 65nm finished CMOS wafer cost these days?

  5. @ The PR says "improved" dark current and full well capacity, not "reduced"...which would not be something to brag about.

    It looks like they revised the PR and fixed the typo. You can see the original version with "reduced" full well in many places on the web, for example here.

    Jokes aside, 1.1um pixel with an acceptable performance is a great achievement and very important milestone for the industry. Howard Rhodes and his team together with S.G. Wuu and his TSMC team did excellent job. My sincere congratulations to them! No PR joke can diminish that success.

    This is the second time that Omnivision and TSMC lead the pixel shrink step. Talking about the history, this is a new turn in IDM vs fabless debate from few years ago. I'm eager to see the performance data of this 1.1um pixel.

  6. Thank you for your kind words.

    The "reduced" joke I have found to be most amusing. The release was originally worded as "improved" but late in the process someone got overly clever and changed it to "reduced" and we missed it. Thanks for your eagle eyes.

    Regarding performance, I am sorry I cannot share more details. Let me say that the performance metrics (QE, crosstalk, SNR10, FWC, dark current, WP) are all very good and stand on their own. I believe customers will accept 1.1u performance. As I stated last year's workshop it is the 0.9 SMP that is the big challenge facing us.

    One note I would like to add to a seperate string of comments. James He is not only an excellent technologist, but a fine human being, completely honest with a keen and deep insight into people whether they come from the West or the East. I am sorry he is no longer with OmniVision and I wish to remain his friend. ... Howard

  7. It is indeed an interesting chapter in the 10-12 year old IDM v. fabless debate. I think there was even an ISSCC evening panel session in 2001.

    I know that OVT/TSMC is usually first with press releases, as they need to be, but who started shipping 1.4 um in volume first? I know other companies have built 1.1 um pixels already without much press so I guess it will be interesting to see who starts selling 1.1 um pixels first.

    Well, besides a long and not always pleasant rivalry between Photobit-Micron-Aptina v. Omnivision, it seems that the OVT/TSMC team is doing some serious heavy lifting on the small pixel engineering challenges. Congratulations to them on this announcement!

  8. @ other companies have built 1.1 um pixels already without much press

    The important difference is that Omnivision is first to present it to the general public in Barcelona. To me this means that the pixel performance has reached a certain level where the company can defend its claims.

  9. Considering his career history, more toleration is required to blog ower to those biased options from E.R.F

  10. @ Considering his career history, more toleration is required to blog ower to those biased options from E.R.F

    The rumor is that Samsung is going to announce its 1.1um pixel-based sensor in Q3 2010, may be even in July. Eric knows what he says.

    Still, Omnivision seems to be the first to demo its 1.1um sensor.


All comments are moderated to avoid spam and personal attacks.