The Back-Illuminated CCD: Improving Light Sensitivity
This article discusses a specialized type of CCD that increases manufacturing complexity but also mitigates performance limitations associated with the more standard approach to CCD implementation.
Welcome to Part 7 of AAC's series on CCD (charge-coupled device) image sensors. In this article, we'll introduce the concept of the back-illuminated CCD.
Before moving on, please consider catching up on the preceding CCD topics below:
- Introduction to Image Sensor Technology
- CCD Structure and Functionality
- Full-Frame, Interline-Transfer, and Frame-Transfer CCDs
- Clocking Techniques for CCD Readout
- CCD Output Signals
- Sampling, Amplifying, and Digitizing CCD Output Signals
What Is a Back-Illuminated CCD?
When we say that an image sensor is front-illuminated, we mean that its physical configuration is analogous to what we expect from a typical IC: pins extend from the perimeter of the device down to the mounting terminals, the “back” of the device faces the surface of the PCB, and the “front” of the device is exposed to incident light.
Most CCDs do indeed operate in this way, but it turns out that performance can be enhanced by physically modifying the device and then mounting it such that incident light arrives from the other side. CCDs that are intended for this type of implementation are called back-illuminated CCDs.
Front Illumination and Reduced Sensitivity
The following diagram, which we saw back in Part 4 of this series, reminds us that CCDs are constructed with a layer of polysilicon gates (aka electrodes) that cover shift registers and potentially light-sensitive portions of the device.
In full-frame and frame-transfer CCDs, the photoactive site is underneath an electrode, such that light arriving from the front face of the CCD must pass through polycrystalline silicon before it can generate electric charge. This causes fairly serious problems, as we’ll see in a moment.
It might appear that the interline-transfer CCD eliminates this issue, because the photodiode is not part of the shift register and therefore doesn’t need to be covered by a polysilicon gate. However, in this case we’re trading one limitation for another: the interline-transfer approach reduces overall sensitivity, since the photoactive sites occupy a relatively small portion of their respective pixels. This effect is mitigated but not eliminated by microlenses that focus light onto the photodiodes.
Light-induced charge generation occurs at different physical depths in silicon depending on something called the absorption coefficient, and the absorption coefficient is in turn a function of wavelength. This phenomenon is conveyed by the diagram below. The red wave represents longer-wavelength photons (such as red light or infrared radiation), and the blue wave represents shorter-wavelength photons (such as blue light or ultraviolet radiation).
In a front-illuminated system, the CCD’s response to light is significantly altered by the presence of polysilicon electrodes. First of all, the electrodes are not perfectly transparent; they scatter and reflect incident light and therefore reduce overall sensitivity.
Furthermore, the electrodes make detection of certain wavelengths basically impossible, because their thickness exceeds the absorption depth. For example, if the polysilicon layer is 500 nm thick, UV radiation with an absorption depth of less than 500 nm cannot produce any electrical response in the CCD.
A CCD’s ability to detect light is conveyed by quantum efficiency (QE), which indicates the percentage of incident photons that are actually converted into usable electric charge. QE can be reported for general optical response or for specific wavelengths. Typical front-illuminated CCDs that are subject to the deleterious effect of polysilicon gates have a maximum QE of approximately 50% near 700 nm and maybe an average of 25–30% across the visible spectrum.
Understanding Back Illumination
If light enters from the opposite side of the device, we completely bypass the troublesome polysilicon electrodes. This increases overall quantum efficiency and is particularly advantageous in applications that require sensitivity to shorter-wavelength radiation.
A back-illuminated CCD equipped with a good anti-reflection coating can surpass 70% average quantum efficiency across the visible spectrum. At certain wavelengths, theoretical values approach 100% and measured values exceed 90%. The following diagram provides a general comparison of front-illuminated QE and back-illuminated QE.
Backside Illumination: The Downside
As you probably guessed, this improved performance comes at a price. First there is the literal price—the thickness of a back-illuminated sensor must be significantly reduced to ensure adequate sensitivity, and this challenging manufacturing procedure makes the device more expensive.
Also, the procedure used to thin the CCD can result in imperfections that increase noise.
It’s also possible that we’ll trade the increased short-wavelength sensitivity for reduced long-wavelength sensitivity, if the CCD is made so thin that the absorption distance of longer wavelengths approaches the thickness of the device.
Near-infrared radiation, for example, might pass right through the silicon substrate. Applications that require IR detection can benefit from a modified form of back illumination in which a thicker substrate is combined with bias voltages that prevent photogenerated charge from being lost via diffusion. These types of CCD sensors are cooled to extremely low temperatures during operation; at normal temperatures there is too much dark current.
We’ve covered characteristics of back illumination and reasons why engineers are motivated to adopt this technology. Back-illuminated CCDs are not the sort of thing that you would put into a cheap camera for making home videos, but they are worth the extra trouble and expense in high-performance systems that need as much quantum efficiency as a sensor can offer.
This is very exciting!! I did not understand how the charge transfer was handled.
I presume one can also put a standing wave into the charge transfer material, and use the phase to shift that in either direction, based on the speed of light in the material and the electron wave function (mainly its velocity). So you could pause it, or speed it up with the phase, keeping each electron bunch in order. I would have to see how much cross talk there is, but it cannot be more than putting mostly passive electrodes on the outside of the charge transfer material where the potential from the array of electrodes is rather coarse. There are many modes to use, and the mode control in laser fibers is pretty welll advanced, so slowing them down to video frequencies should be possible and probably off-the-shelf.
I have an application where the charge transfer is for capacitive and charge sensing and I just need to be able to sample and move the electrons to a readout. Am I making it too complicated? If I have a sensor element generating electrons at varying rates, isn’t that pretty routine?
I need the large arrays of the “optical” sensors, but I do not really need them to be sensitive to light. So is it possible to have a many “floating diffusion” elements manufactured with their own amplifiers, threshold reference voltages, and ADCs?
Is it possible to have an “optical” sensor manufactured without photodetector material at all? It really does make sense. Can it be done? Are there people who will make “photo” detector arrays and readouts to order? I use “photo” arrays now but cover them. The commercial products do not give much control over the reference voltages, amplification and readout. Or I do not know what to ask for. The array manufacturers are so focused on “photo” imaging, that the underlying charge measurement and management is disregarded and messed up, usually to the point that the “photo” sensor is useless for what I want to do.
How I wish I had all your knowledge.
Richard Collins, The Internet Foundation