Thursday, May 08, 2014

Omnivision Explains Image Ghosting Causes

Omnivision's patent application US20140117485 "Negatively charged layer to reduce image memory effect" by Howard Rhodes, Dajiang Yang, Gang Chen, Duli Mao, and Vincent Venezia talks about rarely mentioned but commonly observed ghost image effects: "The electrical signature of an image with high brightness levels that falls onto a complementary metal oxide semiconductor (CMOS) image sensor may remain embedded in subsequently read out electrical signatures of subsequently acquired images. The electrical signature of a previously sensed image remaining in the image sensor has been called a “ghost artifact” or a “memory effect.” This unwanted effect can be exacerbated by repeated exposure of static images, especially high intensity or bright images, to the image sensor. The retention of ghost images represents noise that obscures subsequently acquired images and reduces the signal to noise ratio and may cause blur if there is movement being imaged.

The memory effect problem has been found to be especially present in CMOS image sensors that have been fabricated using advanced fabrication technologies, particularly those employing measures to maximize metal interconnect density. For instance, those fabrication technologies employing so-called “borderless contacts” have been found to be associated with the root cause of this problem.

Omnivision explains: "The deposition of the contact etch stop layer is a fabrication technique that may be utilized when providing borderless contacts, which may be employed to increase metal interconnect density in pixel array."

Contact etch stop layer 322 is deposited over passivation layer 320, which is deposited over the pinning surface layers 313 included in example pixel array 302

"In one example, contact etch stop layer 322 may include a silicon nitride based dielectric including for example, silicon oxynitride, silicon carbide, or the like... The mobile charges in the PECVD silicon nitride and/or silicon oxynitride of contact etch stop layer 322 can be moved by electrical forces such as electrical fields placed across contact etch stop layer 322, which can cause unwanted effects in nearby semiconductor regions, such as photodiode regions 312 and/or the pixel circuitry included in the pixels of pixel array 302. For example, the source to drain resistance of a transistor included in the pixel circuitry included in the pixels of pixel array 302 may be affected by the mobile charge in the overlying PECVD silicon nitride of contact etch stop layer 322 by altering the depletion characteristics of an underlying lightly doped source or drain region.

Furthermore, it is noted that net positive charges can be induced directly in the PECVD silicon nitride and/or silicon oxynitride of contact etch stop layer 322 by exposure to visible light that may pass through contact etch stop layer 322, especially when photodiode regions 312 of pixel array 302 are illuminated with bright light when imaging. In particular, the energy associated with the phonon modes of the Si—Si and Si—H crystal structures may participate in the optical excitation of the electrical carriers. Consequently, memory effect is caused by the generation of positive charges in, for example, the SiON film of contact etch stop layer 322 that overlies the photodiode region 312 under the strong light illumination.

"To illustrate, FIG. 3A shows light 315 illuminating photodiode region 312, which therefore illuminates and passes through contact etch stop layer 322 as shown. This may occur when photodiode region is capturing an image. As a result of this illumination with light 315, positive charge 317 is induced in contact etch stop layer 322, which induces electrons 319 at the surface of the photodiode region 312 surface as shown.

FIG. 3B shows that after light 315 is no longer present and photodiode region 312 images a darker scene or is in a low light condition after having been illuminated with bright light 315 and after the image has been captured, the induced electrons 319 at the surface of the photodiode region 312 are injected into the photodiode region 312, causing the unwanted memory effect. In other words, when the pixel including photodiode region 312 images a darker scene, the induced electrons 319 at the surface of the photodiode region 312 that were a result of the previously captured image are injected into photodiode region 312, which generates localized dark current causing an unwanted “ghost image” of the previously captured image to appear as a memory effect in pixel array 302.

A nice explanation of the positive after-image. Indeed, many CMOS sensors have it positive. In some sensors the after-image is negative though - probably a different effect.


  1. I am having a little trouble accepting this explanation. Perhaps the solution works but the explanation seems unlikely to me. It says that electrons are generated optically at the top of the pinning layer due to "Si-Si" and "Si-H" phonon modes....I guess this means thermally assisted optical generation the interface. Then these electrons hang around in the P+ layer without recombining in this heavily doped material, and then diffuse into the storage well on the next frame. I don't think this happens.

    I would think perhaps radiation during advanced fab causes some deep, slow traps in the lower P-region or N-region that trap electrons and release them later. But, that is just one of many possible hypotheses.

  2. In my experience, the after-images can live a long time, up to few hours, and easily survive multiple power on-off cycles. This kind of time constant is more pointing toward interface or inside-insulator traps. In many cases a change of Tx gate voltage or timing amplifies or reduces after-images, so my thinking was that it's somehow connected to Tx gate operation.

    For example, if photocurrent is high, a part of the hot photoelectrons could be captured in oxide near the Tx drain, that is on FD side, and create a barrier and associated with it slight image lag. Or, in case when the pixel already has a small image lag, its value could be modulated by the light. This can explain a negative after-image.

    If the hot photoelectrons are captured on the PD side of the Tx gate, the image lag can either increase or decrease, depending on the potentials in the area. That could explain a positive after-image.

    1. Seems like another reasonable hypothesis. Charge trapping in the insulator, esp hole traps, can explain hours-long time constants. But do you still think the Omnivision patent app explanation is "a nice explanation of the positive after image"

    2. @ But do you still think the Omnivision patent app explanation is "a nice explanation of the positive after image"

      Their solution has nothing to do with transfer gate and, apparently, works for them. For me, this is the best proof that their explanation is true, in their case.

  3. This hole or electron trapping phenomena can be observed in high dark current semiconducteur materials. High photogeneration can stuff the generation centers by photo electrons-holes. Then the dark current will be lower at these places and result negative after image.

    But in 4T pixel, at normal temperature the dark current is so low that it would be difficult to generate noticeable after-image. There should be other reasons.

    -yang ni

  4. So it is ones again about the interface. Beside this interface generated memory effect in the 4T pixel there is naturally also interface generated 1/f noise, interface generated random telegraph signal noise, interface generated dark noise component in read noise, interface generated fixed pattern noise, and very likely interface generated dark noise. For these reasons in the manufacturing of the 4T pixel one needs to take special care about all process steps affecting the interface quality. Since all these interface related issues can be avoided in a deep buried channel MIG pixel the interface quality has no effect on image quality meaning that image sensor manufacturing is greatly facilitated while at the same time the image quality is improved.


All comments are moderated to avoid spam.