Fudan University's IEEE EDL paper "A Novel One-Transistor Active Pixel Sensor With In-Situ Photoelectron Sensing in 22 nm FD-SOI Technology" by Yong-Feng Cao, M. Arsalan, J. Liu, Yu-Long Jiang, and J. Wan "is commented as “revolutionary” by its reviewer." From the abstract:
"For the first time, a novel active pixel sensor (APS) with 22 nm fully depleted silicon-on-insulator (FD-SOI) technology is experimentally demonstrated. The APS in-situ integrates photo sensing, charge integration, buffer amplification, and random access in one transistor without charge transfer and assistance of additional transistors. The deep depletion effect in the substrate of the SOI MOSFET is used to collect photoelectrons which are then sensed by the top Si channel. A sensor array is proposed and high sensitivity is experimentally verified in a simplified circuit."
IEEE JEDS publishes Peking University, Beijing, open-access paper "UTBB-Based Single Transistor Image Sensor of Submicron Pixel Using Back Gate Modulation" by Liqiao Liu, Xiaoyan Liu, and Gang Du:
"Image sensor has developed for decades. Now, submicron photo sensor device with high performance is required. In this work, a UTBB (ultra-thin body and box) based single transistor image sensor has been investigated. The light collection and signal readout are accomplished by a single transistor, so the pixel of the UTBB image sensor can shrink down to the submicron. The main parameters impacting the performance of the UTBB image sensor such as back voltage, the thickness of the BOX, well doping concentration and well depth are investigated. Besides, the UTBB image sensor can achieve multi-resolution to adapt to different requirements. The performance of the UTBB image sensor is evaluated by TCAD simulations."
The principle is very close to one transistor memory ideas, like Z-Ram (from Innovative Silicon), and other floating-body-charge-modulating-Vt-of-MOSFET principles.
ReplyDeleteIn this case, it's charge in the "floating" back gate (looks like the electron inversion layer is formed under the BOX under illumination - which is essentially floating), but this difference does not invalidate such similarity.
When applied to memories, these ideas failed.
A single cell operates just fine, it is when you make millions of them, their wide Vt distribution functions, the tails of these distributions, is what's making these ideas impractical.
Hundreds of millions of dollars spent by VCs and by large semiconductor companies, all in vain (so far).
It remains to be seen if the idea described in this paper goes from nice basic R&D to production and acceptance.
Interesting work - but misses references to important pioneering work by STM. Arnauld Tournier already worked on 1T-pixels in 2006 - see ICSICT'06, TED'07 Vol 54 and IISW'07
ReplyDeleteThat's 12-13 years ago.
DeleteThe fact that this idea did not take off, and did not take over the world, is an important indicator.
I would say that Olympus famously demonstrated 1T charge modulation devices (CMD) and that is also where the name "active pixel sensor" came from (Tsutomu Nakamura circa 1991). That was soon followed by TI's BCMD (Jerry Hynecek). But, apparently not famously enough.
ReplyDeleteI wonder how the absorption layer (P-) is cleared of electrons on reset? And how is CDS performed? And about 10 other questions. It is always nice to demonstrate new devices. Not sure I would say revolutionary unless I am missing something.
Well if you want really to go back into history, you should check out the work of Brewer and Roks. They did a 1T readout of a CCD, very similar to what is done here with a 1T pixel. I think Brewer was already there in the '70s or '80s, Roks came later.
ReplyDeleteBrewer's CCD output amp was published in 1979. You can find citations to that and many of the 1T APS works in the infamous "dinosaur" paper from SPIE 1993 for those that might not be familiar with those pioneering efforts. E.R. Fossum, Active Pixel Sensors -- Are CCDs Dinosaurs? in CCD's and Optical Sensors III, Proc. SPIE vol. 1900, pp. 2-14, 1993.
DeleteCurrently they are using STI to isolate their pixels, limited to ~300nm depth. So for red light, their resolution will be a lot lower. Switching to DTI will prove difficult as there is no inherent stop layer for the DTI etching.
ReplyDeleteThen there is the issue of influencing the channel. In a conventional readout the electrons are nicely collected on the source follower gate, but here they are "somewhat" guided to be below the channel. Then they have to influence through a much thicker oxide layer than in a conventional source follower.
So, 64200 electrons/lux*s "which is higher than the standard CMOS sensor"... No, standard sensors have SNR, DR, etc. and not just blow up their signal/noise beyond proportion.
Then there is the issue of CDS or DS as Eric mentioned...
Must not be hard to get something published these days ;-)
Agree with your arguments about all the issues if implementing such ideas into a competitive imager, but still, I think, it is not so easy to publish something in one of top journals for electronics as IEEE EDL. And I see their work fits good to the research focus and acceptance criteria for IEEE EDL.
DeleteThe problem is that this monopoly may not provide them those cool stuff one day in the future...
ReplyDeleteI think this just replies the trade war, but still not know why it needs to be involved.
ReplyDeleteIt truly needs to be verified in future study, the performances are not shown in such a short letter
ReplyDeleteI am curious about the fact, that is there any 1T-APS used in a real image sensor product? I guess almost all the image sensor companies are based on the CIS technology.
ReplyDelete