In the beginning of resolution enhancement techniques section, we saw that the resolution is related to the wavelength and numerical aperture. We also saw that depth of focus is affected by the same parameters viz. wavelength and numerical aperture. We also made a note that we need a large depth of focus for a good process.
What is depth of focus?. Consider any of the spectator sports, such as cricket. In some of the sports magazines, sometimes we see the photo of a batsman. Usually the picture is taken from far away (from the boundary line). From the boundary line, the lens is adjusted so that the batsman is in focus. Then the image will be very clear and the resolution will be very good. But when the batsman is in focus in that picture, we can also notice that the bowler and the umpire will not be in focus and they will appear hazy.  The wicket keeper will appear slightly clearer if he stands close to the batsman. So, the distance between the lens and the point of focus (i.e. the batsman) is an important variable.
If we focus the lens exactly on the plane of the batsman, then the resolution will be very good, but the resolution of images of people behind and in front of the batsman in the same picture will not be good . On the other hand, it is possible to take a photo (perhaps with an autofocus camera) which will not be exactly focusing on one plane. In this type of photograph, (for example what we normally take when three or four rows of people are sitting or standing and then you take a photograph) everybody will be visible reasonably well, but none of them will be seen with high resolution.
What this means is that we can take photograph with many different objects (some in the front and some in the back) we can get either (1) all of them in the image with a reasonable image quality, but none of them with very high quality image or (2) we can focus on one object and get a very high quality image, but we will get very poor resolution on the other objects which happen to be in the front or the back. The former is achieved by reducing the lens aperture size and the later is achieved by increasing the aperture size.
How is this relevant for chip production? When we are placing a mask and projecting light through the mask using lenses onto the wafer, we want the image to come accurately on the wafer. What if the wafer is not exactly planar? If the wafer is not planar, (if there are ups and downs) and if the resolution required is very high, the image will not print properly.
To illustrate with another example consider a photographic film. We use light to make photographic reprints. We can make many copies of this and this has to be done in a dark room. The film has to be aligned with a lens and a light. We use a photographic film which is usually a plain paper with a light sensitive chemical coating on top. The film is held stationary, the light is shone through the film and the photographic paper is held straight at the bottom. Then the image  will print correctly. If the paper at the bottom is bent and wavy, then we will not get the correct image.
This is exactly what will happen if the wafer surface is not planar. The wafer surface will never be very planar in the atomic level, but if it is very non-planar (i.e. if there is lot of topography or  ups and downs), then the lithographic process will not print the images correctly. What can be done to correct this problem? We should remove the ups and downs on the wafer and this is done using a process called chemical mechanical planarization or chemical mechanical polishing (CMP).  CMP will help reduce the topography and get a more or less planar surface, but we will never get an ideal atomically flat surface for all the wafers. So the lithography process must be tolerant to some variation in the wafer planarity.
How can we measure the tolerance level to the variation in topography?
One way to do that is as follows: Take a very planer wafer (standard wafer). Take a mask with  a well known pattern  and then place it above the wafer and project the image on to  the wafer. Usually this is done with an auto set up, where the machine itself can find out the exact distance where the focus will be best.  Print the image and this should come correctly.  Next move the lens to the next shot and pull it away from the best focus deliberately by a small distance (0.1 micron or 0.2 micron) and then take the image. In this case, we are deliberately taking it out of focus and continue with the process. Similarly, in the next shot, move the lens towards the wafer (push it towards the wafer) by 0.1 micron and take the image.  In the next row of chips, increase (or decrease) the exposure, while following the same set of focus adjustments. Thus we are adjusting the focus and exposure in a matrix like fashion. This is called focus-exposure-matrix.

Thus the neighboring chips will be made with slightly different focuses and exposures. The remaining processes (such as deposition or etching) to make the lines will be identical. At the end of this sequence, the wafer will be tested to see whether all the features are created correctly. For  example,  if  the best focus itself is not giving the features properly then this process is really poor. But we may find that the best focus as well as 0.1 micron and 0.2 micron out of focus chips can yield good image, but 0.3 micron out of focus yield poor image. Then we can say that if the incoming wafer is not planar up to +/- 0.2 microns then the lithography process can handle that. Thus, the total ups and downs (maximum to minimum) can be 0.4 micron in this case. It also means that if the planarity is poor, (i.e. if the topographical variations are more than 0.4 micron) then we cannot transfer the image from the mask to the wafer successfully. The topography of the wafer has to be reduced by some method before it is sent to lithography.