The basic concept of Wafer Level Chip Scale Packaging (WLCSP) is to employ more or less the same techniques used in making a semiconductor device to create its packaged form. For convenience we will define the device manufacture as having been done in the Front End of the Line (FEOL) and the WLCSP packaging as being done in the Back End of the Line (BEOL). As in the manufacture of the device, the manufacture of the package is done en masse so that economy of scale minimizes cost and it is done in the same format namely on the wafer. There are however differences that arise that creates new needs in the photolithographic processing equipment that performs the Photolithography task. In this Blog we examine some of the most critical of those differences. We begin with the wafer itself.
The Wafer. The value of the wafer at this stage of the process is very high as it has been fully processed and the chips that make up the wafer are at their full value as functional devices. Also because all of the films needed to make a functional device up to that point have been applied the wafer is no longer a nearly perfectly flat object but is rather warped.
The character of the Photoresist to be applied. As the geometries to be created are rather large by comparison to those created in the FEOL the resist tends be far thicker and when bonding bumps are created the photoresist can be used to form a sort of mold into which the desired bump material is plated. In the BEOL resist thickness will typically be on the order of several microns up to 100 microns. The economics of packaging (BEOL) differ from those in the device (FEOL) manufacturing. Because geometries are generally larger lower cost resists can be used and often are. Photoresist processing equipment needs to accommodate these differences and must have a Low Cost of Ownership (Low COO) on a per wafer basis than FEOL tools.
Then the equipment should be able to handle warped wafers, with virtually NO wafer damage, it must be able to use thicker resists and perform the soft bake in an enclosure that does not need to be cleaned frequently if at all and it must be miserly in consumption of photoresist. The equipment itself should have high throughput, low footprint and low cost and material consumption.
Check out our other blogs discussing the BEOL from different points of view.
As it becomes apparent that the adoption of new process nodes has been significantly slowed down due to availability and cost of next generation FEOL lithography technologies, it becomes ever more important that the continuation of cost reductions and functionality improvements that have driven the growth of the Semiconductor industry over the last 40 years be sought elsewhere. Some of the investment heretofore focused on the FEOL “bleeding edge” technology should now be refocused more in BEOL Photo resist lithography for packing where the lower hanging fruit lies. This will require simplification of processes and reduction in material costs wherever possible while at the same time improving process control, yields and functionality and reducing photo resist processing tool costs.
One way to reduce costs is by substantially reducing waste in photolithographic processing technology. It has long been noted by just about anyone who performs spin processing on Photoresist Track equipment that there is an enormous amount of waste in the process as most of the Photoresist material consumed in the process is actually thrown off the wafer during the spin process and wasted. Textbooks in disusing the process point out the wastefulness of the process. The waste becomes especially meaningful in BEOL WLCSP, 2.5D and 3D processing where thicker resist films need to be cast. Multiple spins to achieve the film thickness required exacerbates the waste, while increasing Photoresist viscosity so that the thicker films can be cast adds substantially to equipment costs as the pumps required to pump the material can cost well into the 5 figures in US Dollars. Multiple spins also require more Photoresist track equipment for multiple trips through the equipment. The high viscosities also make elimination of bubbles a constant struggle further increasing costs. The film is formed during the spin process and then baked typically on a hot plate to remove additional solvent to render the film stable and photo definable. All of the foregoing tends to substantially increase photo resist processing costs in the Back End of the Line (BEOL) lithography for Wafer level Chip Scale Packaging (WLCSP). Another avenue is to use expensive chemically amplified resist (CAR) systems, which have the effect of improving CD control and aspect ratio’s in a single coat, however further increasing material cost substantially.
The engineers at S-Cubed™ have invented a new process…FastSpin™ that addresses all of these cost/complexity issues. In the new process (patents applied for) the film is not cast so much as wet formed by spinning the wafer under exquisite servo control for a very short time, subsequently processed to render the film stable and photo definable. The wet film is maintained in the wet state and is allowed to level, allowing for planar coatings over topography. The process not only substantially reduces photo resist consumption for both standard and CAR resists, it also substantially reduces the cost of the photo resist processing tools, Photo resist track, by reducing the number of spin processors required for a given spin processor tool throughput.
Please request further information regarding S-Cubed™ FastSpin™ click on the link:
The Semiconductor Industry has for years followed the path of “Moore’s” Law. Gordon Moore’s forecast that the industry would every 18 months double the number of transistors in a given area, thereby increasing value while reducing cost. His prediction has held true for about 20 years, and is now faltering. However the effect of his prediction can and will continue into the foreseeable future but rather than the aerial rule obtaining on silicon it will now move increasingly to the package. That trend is now beginning in earnest and the improvements in functionality and cost will occur more and more in the Back End of the Line (BEOL) /packaging of function. The driving economic factor is clearly the proliferation of mobile devices of all sorts.
While the technology of the photolithographic processing remains very similar there are important details that must be addressed if the tool sets of track and lithography are to be optimized for the new needs. BEOL processes involve different economic drivers, which must be accommodated by the Cost of Ownership (COO) models for BEOL tool sets versus Front End of Line (FEOL) tool sets. For more detail on the subject please see our BLOG entitled “Why Back End of Line (BEOL) Photoresist “Track” tool are and must be different from Front End of Line (FEOL) Photoresist processors.”
“No tree grows to the sky”, my friend Karol Ubanek, the founder of Tencor, uttered these words of wisdom. They come to mind whenever I read an article or report on the subject of EUV. The amount of money and effort that are being poured into the development of the technology is nothing short of astonishing. One can only wonder what would happen if so much treasure was lavished not on the continuation of yesterday (smaller wavelengths equals smaller critical dimensions) but rather on totally tomorrow and different approaches to the problem of the continuation of the eternal “shrink” in semiconductor production. What if the answer wasn’t in the shrink but somewhere “outside the box”? TSV is such an attempt and so is 2.5 D packaging, occasionally referred to as 3.5D. . Maybe virtual wafers of epoxy rather than silicon eWLB style is the answer and then again maybe we are all trying to answer the wrong question. Is the question really to pack more devices into fewer square millimeters or is it really a better idea to “roll up the devices” as in a rolled up carpet, (there is a lot of visual information in a little bit of volume when the carpet is rolled up…just an example). It just seems to me if one half the money that is being spent on EUV was spent on seeking new questions and maybe getting new answers it could be more productive, after all… indeed, “no tree grows to the sky”.
A very important aspect of cost of ownership (COO) of tools that include spin coaters and developers (often referred to a photoresist track tools) in the back end of line (BEOL), MEMS, and patterned sapphire substrate (PSS) processing-type processing lines is one that does not appear on the COO software packages. Often referred to as “opportunity cost,” it is specifically defined as “the difference in return between an investment one makes and another that one chose not to make.” A high degree of system “granularity” enables appropriate investment as customer needs evolve. Ultimately this can substantially reduce opportunity cost. Granularity specifically relates to tools sized to the capacity need, as well as properly balanced for output in order to minimize waste. Lets look at how that works, in the context of opportunity cost.
Since opportunity cost is always a comparison of the path chosen to the path that was not chosen, we will compare two hypothetical systems whose COO on a per wafer output basis is identical, but its granularity is different. Granularity will tend to go up in integer values. Let us suppose a system whose out put is equal to one unit of capacity, (i.e., 2000 wafers per month) and another whose out put is equal to two units of capacity (i.e., 4000 wafers per month). If the need is for one unit of capacity, then if the COO is the same and the more granular system is the clear choice. If it is two units of capacity then the choices are equal. However what happens when it is three units of capacity; then four; then five? Capacity needs often develop in incremental steps over the long term instead of in immediate doublings. The investment in the “granular approach” is obviously the more appropriate forward-thinking plan. As needs eventuate, so that capacity can be adjusted to market needs at the lowest possible cost in capital spread over time.
In our first post, we made the following point about companies that are not the huge Foundries; “such companies will use their own creativity and drive to gain a market advantage owing to their expertise in both designing and building their chips. Such companies will need to find ways to lower their manufacturing costs by effectively lowering the cost of ownership of the tools of production.” Another way to make this argument is that those companies must innovate.
But wait. Everyone knows that we all need to be data driven. We need to have the data that successfully makes the argument that the innovation is working successfully. It is sort of a “chicken and egg” dilemma. If we had the data, then it’s been done and if we don’t have the data we can’t proceed. Ultimately, we have to start someplace. The question is always “where?”
First, we need to start with a better understanding of nature. When all else fails, we as professionals should be thinking through what nature wants to happen. The study of chemistry and the laws of thermodynamics in the discipline of physical chemistry can, and should, instruct us and provide the bridge that we build to cross from no data to innovation; and from innovation, to new data that can provide that lower cost of ownership so vital to the companies that must be creative. Ultimately, it is that creativity that helps to foster success.
In earlier writings we have discussed cost of ownership issues (COO) related to photoresist processing tools. In this entry, we will discuss the important functional differences between tools generally intended for the FEOL, and those more adaptable to the needs of BEOL wafer fabrication.
As the semiconductor industry has packed more and more functionality and speed into ever smaller chips (here one almost always must refer to Moore’s Law well know to those in the industry), the size and cost of the packaging of the chip has made necessary the creation of new types of wafer processing, which is needed to do wafer-level chip scale packaging. In this specific technology, photolithographic processing is used so chip outputs and inputs can be repositioned, so as to adapt to interconnect circuitry directly without the intervention of a typical pin out type package. In this case the chip is “rewired” and “bumped” so that it can be “flipped and bonded” directly to the chip interconnect circuitry. Such rewiring and wafer bumping is accomplished on the finished device wafer prior to singulation of the wafer into chips.
While the patterning concept is identical to that required to make the chip itself in practice, the geometries are far larger and the need for organic insulating film more prevalent than in the creation of the chip. This means that the spin cast films are much thicker and there is far more effluent of material leaving the wafer during the required “bakes” done in the thermal modules of the requisite tools. These thicker films require the following considerations in the design and implementation of the tool.
- The cost of the process must be kept low, thus requiring tools with a low COO.
- The spin cup must be designed to handle thicker films and capable of modulating spin exhaust conditions on a step-by-step basis throughout the process recipe.
- The higher viscosity resists that are used to spin the thicker films means pumps must be geared to accomodate those viscosities.
- The higher viscosity resist requires that the dispense height be preferably programmable rather than adjustable so that it can be varied during the dispense.
- The much thicker films evolve far more solvent and other effluent requiring that the effluent be directed away from cold surfaces where it can condense and require frequent bake module cleaning, or create defective films owing to effluent reflux.
While FEOL tools may be utilized in BEOL, their high cost can be prohibitive, and their performance while tuned to the FEOL is not appropriate to the BEOL provider.
In an earlier BLOG we discussed “Lowering the Cost of Ownership (COO) of Spin Process Tools.” we did not discuss the fact that the customer does have an option to purchase a used, “used and refurbished” or “as is” version of a tool built at some point in the past as another means of lowering cost.
There are some obvious and immediate COO issues that come to mind when used equipment, sometimes more respectfully referred to as “legacy” equipment is purchased as opposed to new equipment. They include:
- Very short warranty. Used or legacy equipment typically has a 30 to 90 day warranty, whereas today’s new equipment will often have a two-year warranty on components and one year on labor.
- Availability of spares. Used equipment, some of which was built up to 30 years ago, is becoming more and more difficult to support as the electronic control means are often no longer in production. While this issue is mitigated when new controls are implemented on old platforms, however this adds to the overall cost.
- Foot print issues. Equipment made before the advent of process stacking, typically done on the thermal modules, occupies more than twice the space as modern tools with Stacking. Used equipment that has stacking capability is often more expensive than new tools with stacking.
- Installation Costs. Frequently it can cost as much to install a used tool as it cost to buy it. That is not the case with new equipment, where installation costs and facility costs are low.
In addition to the direct cost factors noted above, there are more subtle factors that come into play. Including:
- Process support. New tools come with factory process support not typically available in used tools.
- Software support. New tools can be “tweaked” to meet specific customer requirements, whereas used tools typically are not available with ANY software support or modification.
- Capability to meet new process requirements that did not exist at the time of the design and manufacture of the used tool. Such factors as multiple movable dispenses, high reliability robots, hot plates designed to handle thicker photoresist films typically used in special purpose photolithography.
- A great deal of the used “track” equipment was built before the advent of Semi Safety Standards and is difficult, if not impossible, to render safe.
The photoresist process equipment available today is low in cost and high in functionality and reliability. It can be tailored to the specific needs of the very specific customer who buys new and not used.
In our last BLOG we discussed the need to lower the COO of the tools of production for those who are both designing and building their chips. The issue that we attempt to address here is how does one make that happen and at the same time improve functionality permitting the customer to do things that perhaps his competitor is not even aware of as a problem.
The aspects (solid dots below) and the responses (open dots below) to COO issues are:
- The cost of the Bill of Materials (BOM) of the tool.
- Employee identical hardware throughout the product line to improve economy of scale.
- Minimize non-productive hardware, sensors and interconnect complexity.
- Do not compromise on component quality, minimize the number of components.
- Minimize Tool Footprint
- Stack processes where it makes sense to do so
- Integrate wafer handling and processes to eliminate wasted space
- Maximize Reliability (note the very same responses to minimizing BOM costs improve reliability a clear Win-Win.
- Employee identical hardware throughout the product line to improve economy of scale. (Enables testing and long term improvement)
- Minimize non-productive hardware, sensors and interconnect complexity. (components that are not there cannot fail)
- Do not compromise on component quality, minimize the number of components. (The highest quality components fail less frequently)
- Minimize consumption of materials cost. (Things like photo resist, solvents, gases, etc.)
- Improve tool functionality while lowering BOM cost
- Creatively work with customers to take advantage of improved tool functionality.
- Use software to improve functionality as well as hardware. Software has zero replication cost.
- Increase Tool Throughput
- Provide wafer handling capability at low cost that balances process times with minimum handling overhead time. Provide smart robotics capability.
Use “Small Grain” tools. By this we mean tools that can be provided in small increments of production while at the same time minimizing COO and Investment in absolute terms. All of the above items support this Aspect.
Does the word “commodity” come to mind; how about the word “innovation?” OK, now try “risk.” The question of the outlook for semiconductor production worldwide is a mixture of the feelings and fears, thoughts and challenges associated with the relationship between those three words, Commodity, Innovation and Risk…
Now look at the Global Semiconductor Industry and the rise and subsequent consolidation of the Foundries that now dominate the latest mass production semiconductor technologies. There are three possibly four left standing, TSMC, Global Foundries, Samsung and perhaps Intel will enter the battle. The risks of technological advance in the manufacture, as opposed to design, of semiconductors are being born by the foundries. Their hard earned and costly manufacturing skills will be generally available to all who pay the price and have the manufacturing volumes that justify the costly tooling and sustaining engineering.
What of those, who for good reasons of their own, will not or cannot utilize the giant foundries. Such companies will use their own creativity and drive to gain a market advantage owing to their expertise in both designing and building their chips. Such companies will need to find ways to lower their manufacturing costs by effectively lowering the cost of ownership of the tools of production.