Friday, November 30, 2007
Omnivision Lost HP PC Camera Order to Micron
Digitimes: HP has chosen Micron over Omnivision as a secondary CMOS sensor supplier as Micron is offering more competitive prices, according to sources close to Omnivision. Losing the orders to Micron does not suggest the solutions from Omnivision are inferior in quality, it was simply due to the aggressive pricing that Micron offered, the sources stressed. The order amount is not large, which implies any corresponding impact on OmniVision will be limited.
Omnivision Conference Call Minutes
Seeking Alpha published Omnivision's Q2 FY2008 conference call transcript.
Some interesting points from the call:
Some interesting points from the call:
- 1.75um pixel image quality is substantially better than the prior 2.2 micron pixel generation.
- Omnivision believes it's ahead of the competition in the development of 1.4um pixel.
- Mobile handsets accounted for approximately 65% of the Q2 revenues.
- The 2MP and higher resolution sensors approached 15% of total units last quarter.
- 1/4" 3MP is winning sockets at a very fast pace, including multiple tier 1 design wins.
- The total market for sensors in the notebook business will exceed 25M units in 2007, representing more than 25% share of all notebooks.
- TSMC remains by far and away the biggest supplier and Omnivision does not expect that to change at any time in the future.
- Powerchip is a development program which was originally intended and still is planned to deliver product in the second half of 2008. It's too early to tell how this will work.
Hynix History
SIKOD presents Hynix history - a mix of ups and downs. Ups are attributed to amasingly good business practice, while the government helps it to survive during the down periods.
Omnivision's Margins Improving
Yahoo: Omnivision reported its margins improving in Q2 FY2008. Gross margin for the second quarter of fiscal 2008 was 25.2%, compared to 23.4% for the first quarter of fiscal 2008. The increase in gross margin reflects lower production costs and improved yields.
Omnivision reported acceleration of demand for higher resolution sensors for developed markets, expansion in new markets and Emerging Products business improving.
Omnivision reported acceleration of demand for higher resolution sensors for developed markets, expansion in new markets and Emerging Products business improving.
Wednesday, November 28, 2007
Image Sensors on ISSCC 2008
ISSCC 2008 published an advance program. It contains quite a lot image sensor related stuff.
First, there is Imager Design Forum, almost entirely devoted to high dynamic range solutions.
Image sensors technology session has 10 papers. Probably the most interesting among them is one written by Keith Fife:
A 3MPixel Multi-Aperture Image Sensor with 0.7μm Pixels in 0.11μm CMOS
K. Fife, A. El Gamal, H-S. Wong
Stanford University, Stanford, CA
A multi-aperture imager sensor is designed to reduce lens requirements, produce 3D maps and improve pixel-defect tolerance. It comprises a 166×76 array of 16×16 0.7μm full-frame transfer CCD sub-arrays, a CMOS readout circuit and per-column 10b ADCs fabricated in a 0.11μm CMOS process. Snap-shot image acquisition with CDS is performed at up to 15fps. The array has 0.15V/lx·s sensitivity, 3500e- full-well capacity, 5e- read noise, 25e-/s dark signal, 57dB DR and 35dB peak SNR.
The most surprising one comes from Toshiba:
A White-RGB CFA-Patterned CMOS Image Sensor with Wide Dynamic Range
E. Yoshitaka
Toshiba, Yokohama, Japan
A 2Mpixel CIS for mobile imaging applications has a pixel pitch of 2.2μm and is fabricated in a 0.13μm CMOS technology. The sensor uses a white-RGB color filter array instead of the regular Bayer pattern to improve the low-light SNR by about 3dB. The array also achieves a wide dynamic range by using charge skimming and multiple acquisitions. The dynamic range can be expanded by 8 while maintaining the improved SNR.
I bet nobody expected that Toshiba would follow Kodak path in exploiting the modified Bayer pattern with white pixels. In the meantime Kodak presents a hole-based photodiode:
Low-Crosstalk and Low-Dark-Current CMOS Image-Sensor Technology Using a Hole-Based Detector
E. Stevens(1), H. Komori(2), H. Doan(1), H. Fujita(1), J. Kyan(1), C. Parks(1), G. Shi(1), J. Wu(1)
(1) Eastman Kodak, Rochester, NY, (2) Eastman Kodak, Yokohama, Japan
A CIS with a hole-based pinned photodiode is presented. The detector reduces crosstalk by 3× and dark current up to 5× compared with an equivalent electron-based pinned photodiode detector. This technology is capable of full-well capacities of 60kh, 11kh and 4kh for 4.3μm, 1.75μm and 1.4μm pixels, respectively. A red-into-green pixel crosstalk of 7% is achieved at 650nm for a 4.3μm pixel, and the measured dark-current density is 25pA/cm2 at 60°C.
Another noteworthy paper comes from Matsushita on high DR sensor:
A 140dB-Dynamic-Range MOS Image Sensor with In-Pixel Multiple-Exposure Synthesis
T. Yamada, S. Kasuga, T. Murata, Y. Kato
Matsushita Electric Industrial, Takatsuki, Japan
A wide dynamic range 177×144pixel CMOS image sensor that can simultaneously capture both dark and bright objects in one synthesized frame has an analog accumulator within each 8μm pixel to synthesize a wide dynamic range image from multiple exposures. The sensor is capable of acquiring a 140dB dynamic range image at 15fps without external frame buffers.
Another hot topic is random telegraph noise and ways to reduce it:
A CMOS Image Sensor with a Buried-Channel Source Follower
X. Wang(1), M. Snoeij(1,2), P. Rao(1), A. Mierop(3), A. Theuwissen(1,4)
(1) Delft University of Technology, Delft, Netherlands, (2) Texas Instruments, Erlangen, Germany, (3) DALSA Semiconductor, Eindhoven, Netherlands, (4) Harvest Imaging, Bree, Belgium
A CIS fabricated in a 0.18μm process with a 4T pinned-photodiode pixel and a buried channel source follower (BSF) is presented. Measurements confirm that the BSF reduces the dark random noise by more than 50% and improves the output swing by almost 100% when compared with a surface-channel NMOS source follower. Moreover, the BSF reduces the variance of the dark random noise distribution by minimizing the number of pixels that have RTS noise.
Life sciences session has a paper on eye-implantable image sensor for blind people:
CMOS Imager Technologies for Biomedical Applications
J. Burghartz, T. Engelhardt, H-G. Graf, C. Harendt, H. Richter, C. Scherjon, K. Warkentin
IMS CHIPS, Stuttgart, Germany
Two CMOS imager chips are described. The first is a sub-retinal implant, being the only imager chip ever implanted into a human eye that partially restores vision to a blind patient. The second is a miniature imager chip, based on a thin-film-on-CMOS (TFC) pixel technology provides an optimum trade-off between sensitivity and pixel size.
First, there is Imager Design Forum, almost entirely devoted to high dynamic range solutions.
Image sensors technology session has 10 papers. Probably the most interesting among them is one written by Keith Fife:
A 3MPixel Multi-Aperture Image Sensor with 0.7μm Pixels in 0.11μm CMOS
K. Fife, A. El Gamal, H-S. Wong
Stanford University, Stanford, CA
A multi-aperture imager sensor is designed to reduce lens requirements, produce 3D maps and improve pixel-defect tolerance. It comprises a 166×76 array of 16×16 0.7μm full-frame transfer CCD sub-arrays, a CMOS readout circuit and per-column 10b ADCs fabricated in a 0.11μm CMOS process. Snap-shot image acquisition with CDS is performed at up to 15fps. The array has 0.15V/lx·s sensitivity, 3500e- full-well capacity, 5e- read noise, 25e-/s dark signal, 57dB DR and 35dB peak SNR.
The most surprising one comes from Toshiba:
A White-RGB CFA-Patterned CMOS Image Sensor with Wide Dynamic Range
E. Yoshitaka
Toshiba, Yokohama, Japan
A 2Mpixel CIS for mobile imaging applications has a pixel pitch of 2.2μm and is fabricated in a 0.13μm CMOS technology. The sensor uses a white-RGB color filter array instead of the regular Bayer pattern to improve the low-light SNR by about 3dB. The array also achieves a wide dynamic range by using charge skimming and multiple acquisitions. The dynamic range can be expanded by 8 while maintaining the improved SNR.
I bet nobody expected that Toshiba would follow Kodak path in exploiting the modified Bayer pattern with white pixels. In the meantime Kodak presents a hole-based photodiode:
Low-Crosstalk and Low-Dark-Current CMOS Image-Sensor Technology Using a Hole-Based Detector
E. Stevens(1), H. Komori(2), H. Doan(1), H. Fujita(1), J. Kyan(1), C. Parks(1), G. Shi(1), J. Wu(1)
(1) Eastman Kodak, Rochester, NY, (2) Eastman Kodak, Yokohama, Japan
A CIS with a hole-based pinned photodiode is presented. The detector reduces crosstalk by 3× and dark current up to 5× compared with an equivalent electron-based pinned photodiode detector. This technology is capable of full-well capacities of 60kh, 11kh and 4kh for 4.3μm, 1.75μm and 1.4μm pixels, respectively. A red-into-green pixel crosstalk of 7% is achieved at 650nm for a 4.3μm pixel, and the measured dark-current density is 25pA/cm2 at 60°C.
Another noteworthy paper comes from Matsushita on high DR sensor:
A 140dB-Dynamic-Range MOS Image Sensor with In-Pixel Multiple-Exposure Synthesis
T. Yamada, S. Kasuga, T. Murata, Y. Kato
Matsushita Electric Industrial, Takatsuki, Japan
A wide dynamic range 177×144pixel CMOS image sensor that can simultaneously capture both dark and bright objects in one synthesized frame has an analog accumulator within each 8μm pixel to synthesize a wide dynamic range image from multiple exposures. The sensor is capable of acquiring a 140dB dynamic range image at 15fps without external frame buffers.
Another hot topic is random telegraph noise and ways to reduce it:
A CMOS Image Sensor with a Buried-Channel Source Follower
X. Wang(1), M. Snoeij(1,2), P. Rao(1), A. Mierop(3), A. Theuwissen(1,4)
(1) Delft University of Technology, Delft, Netherlands, (2) Texas Instruments, Erlangen, Germany, (3) DALSA Semiconductor, Eindhoven, Netherlands, (4) Harvest Imaging, Bree, Belgium
A CIS fabricated in a 0.18μm process with a 4T pinned-photodiode pixel and a buried channel source follower (BSF) is presented. Measurements confirm that the BSF reduces the dark random noise by more than 50% and improves the output swing by almost 100% when compared with a surface-channel NMOS source follower. Moreover, the BSF reduces the variance of the dark random noise distribution by minimizing the number of pixels that have RTS noise.
Life sciences session has a paper on eye-implantable image sensor for blind people:
CMOS Imager Technologies for Biomedical Applications
J. Burghartz, T. Engelhardt, H-G. Graf, C. Harendt, H. Richter, C. Scherjon, K. Warkentin
IMS CHIPS, Stuttgart, Germany
Two CMOS imager chips are described. The first is a sub-retinal implant, being the only imager chip ever implanted into a human eye that partially restores vision to a blind patient. The second is a miniature imager chip, based on a thin-film-on-CMOS (TFC) pixel technology provides an optimum trade-off between sensitivity and pixel size.
DxO Offers ISP Solutions
Yahoo: DxO Labs announced three embedded image processing solutions for camera phones, available as silicon IP.
The three are the DxO IPE, DxO ISP and DxO DOP families. The DxO IPE family of solutions combines image processing with enhanced depth-of-field. The DxO ISP family provides image processing without the enhanced depth-of-field; and the DxO DOP family provides the enhanced depth-of-field without the image processing functions.
All of the solutions are implemented using configurable and programmable SIMD processor core - sounds a lot like like NXP product. The solutions support resolutions from 1.3MP to 12MP. They are available to be embedded on CMOS sensor, on companion chips inside camera modules, or on baseband or application processor chips.
The three are the DxO IPE, DxO ISP and DxO DOP families. The DxO IPE family of solutions combines image processing with enhanced depth-of-field. The DxO ISP family provides image processing without the enhanced depth-of-field; and the DxO DOP family provides the enhanced depth-of-field without the image processing functions.
All of the solutions are implemented using configurable and programmable SIMD processor core - sounds a lot like like NXP product. The solutions support resolutions from 1.3MP to 12MP. They are available to be embedded on CMOS sensor, on companion chips inside camera modules, or on baseband or application processor chips.
Hynix to Start CIS Mass Production in Q4'2008
The Korea Times: Hynix plans to start mass-producing the CIS products during the fourth quarter of 2008.
By that time it should base its products on 1.4um pixel, if it wants to be on the forefront of the technology. To me, its hard to believe that Hynix is able to develop it in a so short time. If not, Hynix would compete on price with "me-too" 1.75um sensors.
A year ago ProMOS had announced similar plans. However, it clearly takes them more than a year to come up with real products.
By that time it should base its products on 1.4um pixel, if it wants to be on the forefront of the technology. To me, its hard to believe that Hynix is able to develop it in a so short time. If not, Hynix would compete on price with "me-too" 1.75um sensors.
A year ago ProMOS had announced similar plans. However, it clearly takes them more than a year to come up with real products.
Tuesday, November 27, 2007
Hynix Partners with Siliconfile
Digitimes: It's official now. As many have predicted, Hynix signed a partnership agreement with Siliconfile. With all its recent advances, Siliconfile is probably too expensive for troubled Hynix to acquire outringht. So Hynix will acquire just a portion of Siliconfile's equity to strengthen the relationship.
Hynix will manufacture and sell CMOS sensors using SiliconFile's design and will provide foundry services to SiliconFile. Both parties plan to cooperate on product development thereafter.
Siliconfile projects that its annual sales to hit US$65 million in 2007, up from US$24 million in 2006. Siliconfile holds approximately 8% share of the mobile CMOS sensor market. It employs more than 40 experienced engineers, as Hynix said. It's impressive achievement for such a small team to capture so large market chunk.
There are clear winners and losers in this marriage. The most immediate loser is Dongbu, the former Siliconfile's foundry. Dongbu has been with Siliconfile since its inception in 2002 and now it has lost its only big CIS customer. So Dongbu left with a significant knowledge in pixel design and ready to use image sensor IP blocks, like ISP. One can expect Dongbu would try to find some use for it.
Hynix has found the best fit for its CIS program. I'm sure that Siliconfile will propel Hynix to the first league of image sensor manufacturers. However, it might take a year or so for technology transfer. It's probably too late for Hynix to enter to 1.75um pixel generation. I wonder whether Hynix-Siliconfile can provide a competitive advantage in 1.4um pixel.
Siliconfile on its side can enjoy much more advanced process capability of Hynix, so its chances to design a competitive 1.4um pixel are not negligible.
Hynix will manufacture and sell CMOS sensors using SiliconFile's design and will provide foundry services to SiliconFile. Both parties plan to cooperate on product development thereafter.
Siliconfile projects that its annual sales to hit US$65 million in 2007, up from US$24 million in 2006. Siliconfile holds approximately 8% share of the mobile CMOS sensor market. It employs more than 40 experienced engineers, as Hynix said. It's impressive achievement for such a small team to capture so large market chunk.
There are clear winners and losers in this marriage. The most immediate loser is Dongbu, the former Siliconfile's foundry. Dongbu has been with Siliconfile since its inception in 2002 and now it has lost its only big CIS customer. So Dongbu left with a significant knowledge in pixel design and ready to use image sensor IP blocks, like ISP. One can expect Dongbu would try to find some use for it.
Hynix has found the best fit for its CIS program. I'm sure that Siliconfile will propel Hynix to the first league of image sensor manufacturers. However, it might take a year or so for technology transfer. It's probably too late for Hynix to enter to 1.75um pixel generation. I wonder whether Hynix-Siliconfile can provide a competitive advantage in 1.4um pixel.
Siliconfile on its side can enjoy much more advanced process capability of Hynix, so its chances to design a competitive 1.4um pixel are not negligible.
Monday, November 26, 2007
Omnivision Gets New Foundry Agreement
Robert W. Baird reports that Omnivision got a more favorable foundry agreement. Anybody knows what is this new agreement? Has Omnivision improved its contract with TSMC or switched over to other foundry?
Thursday, November 22, 2007
Samsung Develops 150um Pixels
Yahoo: Samsung Mobile LCD Division has completed development of a flat panel X-ray detector (FPXD) for radiology machines using digital imaging with thin-film transistor (TFT) technology. To create its FPXD, Samsung attached photodiodes to a TFT substrate that was produced using its proprietary amorphous silicon technology. The X-rays are detected photon by photon and then converted into visible light, which in turn is converted to electrical signals.
Samsung has created an image enhancement function which is claimed to eliminate most digital image noise interference to provide the highest radiographic sensitivity in the industry.
The new FPXD measures 45 centimeters (cm) wide x 46cm high (or 61cm diagonally) and boasts a 3072 x 3072 (9.4 megapixels) resolution providing ultra-high definition images. The device will be available worldwide beginning first quarter, 2008.
Samsung has created an image enhancement function which is claimed to eliminate most digital image noise interference to provide the highest radiographic sensitivity in the industry.
The new FPXD measures 45 centimeters (cm) wide x 46cm high (or 61cm diagonally) and boasts a 3072 x 3072 (9.4 megapixels) resolution providing ultra-high definition images. The device will be available worldwide beginning first quarter, 2008.
Subscribe to:
Posts (Atom)