Cost Reduction Re-Design
Charles Eidsness
In the course of my career I've been fortunate enough to be involved in several Cost-Reduction Re-Design (CR) Projects. This article shares some advice that I've picked up along the way, but first I'd like to give a little sales pitch on CR Projects themselves.
Cost-Reduction is Fun (by Tom Sawyer)
For some reason CR Projects seem to have a bad reputation among Hardware Designers. I guess they're not viewed as being sexy and cutting edge enough but of all of the projects that I've been involved in I've found CR Projects to be the most rewarding. You can judge your contribution to your employer directly, i.e. if you re-design a card and save $500 per unit and they sell 50000 units a year you've just saved your employer 25million dollars a year.
Arguably cost-reduction re-design is the most important design activity for the long-term health of a company, and the lack of cost reduction re-design is one of the main reasons that the Telecom Industry in particular is doing so poorly. Once you have established customers on an established product every penny you can squeeze out of that product is bonus profit.
The cost of performing a CR Re-Design is usually smaller that the cost of performing a New Product Design and unlike fancy new products there is minimal market risk because you are already reasonably certain that you're going to sell this product. As anyone who has worked in the "High-Tech" industry knows very few new products actually get to market. As an Engineer it can be disheartening to have to throw design after design into the trash. As a Business it can it can be infuriating. Cost-Reduction is something that you can invest X amount of dollars in a be fairly certain you will get Y dollars back at the end of the day. CR Projects almost always go to production. The Telecom and Defense Industries for some reason have failed to grasp this concept. If I was CTO of one of these companies the first thing I would do is dedicate at least 30% of my R&D budget to Cost-Reduction redesign. In fact I'm amazed that these big corporations continually cut operating costs, i.e. employees when they could re-target at least some of those resources to Cost-Reduction Re-Design, ensuring the long-term stability of the company. Any company's high volume products should be going through a continuous cost-reduction cycle every 16-24 months or so. The Commercial Electronics Industry seems to have it right (because of higher volumes I suppose). For instance, every year or two a new Play-Station 2 comes out which represents a significant cost-reduction on its predecessor.
The other thing that I enjoy about CR Projects is that you already have a set of requirements, and where those requirements are lacking you can reverse engineer the current design to generate them (though this can be dangerous and should be done intelligently as it assumes the current design is perfect, which is rarely the case) The "Requirements Two-Step" between the Systems and Hardware Designers is eliminated.
A final thing that I've really enjoyed about Cost-Reduction Re-Design is the challenge. When you're starting with a clean-state on a new product you can architect the design in a way that minimizes design difficulty (at least you should). Generally you confront similar issues and solutions over and over again. "Text-book"-type stuff but on CR Projects you are not starting from a clean-slate. You will face unique challenges, the kinds of challenges that you won't find solutions to in any text-books. The kinds of challenges that you have to be creative to solve. For instance, on one CR Project that I was working on we were upgrading several FPGA's to newer, lower-cost technology but we didn't have the source code/compiler for the software running on the processor and one of the FPGA's that was being changed was the SDRAM Controller. Needless to say there were some bugs in the new controller and the card wouldn't boot. I ended up having to reverse engineer the boot code using a logic analyzer and disassembler to figure out were it was crashing and what was going wrong. It was tedious but a great learning experience and something that I would have never seen on a new product.
Cost-Reduction Proposal
The first place to start on a CR Project is with the Proposal. It allows the Designers and Management to have discussions about risk, cost, etc and once it has been parsed into a single set of options it can be used as a blue-print for the Design activity. I usually divide up a proposal into three sets of options. Like the 3 Bears I have one set with minimal changes/risk, one section with a balance of risk and changes and finally a very aggressive section that attempts to produce a design at absolute minimal cost. This section will almost never be implemented in its entirety but going through the process of attempting to come with a minimum cost design may generate other lower-risk/cost items that I wouldn't have thought of otherwise. Usually I will end up implementing Option 2, with some items shuffled around depending on the preferences of the rest of the Design and Management Group.
In each of the three sections I list out each proposed change including block diagrams, current cost, expected cost savings, time/cost of design activity and risk (on a scale from 1 to 3). A single proposed change can be in multiple sections. In fact usually each section is essentially a subset of changes in the next higher risk section. I also include a summary of total expected cost savings and development cost for each section. I generally include a healthy amount of margin in my estimates since there are still so many unknowns but I don't want to add so much margin that the project doesn't appear to be financially viable the project can't get its feet off the ground. It's a balancing act.
One important question that has to be answered early is "Will we have S/W Development Resources?". S/W is expensive to develop and most companies would prefer a CR Re-Design that doesn't include software changes. Even if it's decided that there will be no S/W changes I usually include bonus proposals with S/W Changes for possible future inclusion. Or maybe someone will change their mind when they see there is a significant savings available. FPGA Development can also be expensive but it is generally worth it (discussed later) but still, make sure FPGA Development costs are properly accounted for in the proposal
Design Review
I don't start writing a new CR Proposal without the Current Design's Schematic, Layout, Design Docs, Test Reports, and a BOM including component costs. The BOM is critical, you can't do a CR Project if you don't know current cost. I start by reviewing the schematic and layout like I would review someone else's design for a design review. I highlight any perceived technical risk in the current design. Quality improvement is another selling point for re-designing a product. Make sure to highlight any potential design improvements in the proposal. I will also highlight any areas of the design that don't have adequate design data. They will be more difficult to modify, like discrete designs with no notes, simulations, etc., or memory interfaces with no timing analysis, or lack of test coverage, etc. If the original design was not done properly the cost and risk of the CR Project will be greater, especially if you are keeping sections of the current design. I also take a look at the layout to see if it was done in a way that makes it possible to rip up areas without effecting other unrelated sections. If the design was auto-routed for instance you will not be able to make changes without ripping up everything and re-routing it (hopefully the original constraints were well defined and available). The state of the current design will dictate the cost of making changes. I almost always end up using the current design's block diagrams and requirements only and throw away most of the schematic and the layout. If you're making more that a few changes it's usually cheaper but it's a decision that has to be made on a project by project bases.
After reviewing what's there I look at the BOM. I sort it based on cost. You will usually find that about 10% of the components make up about 90% of the BOM cost. The top 10% are the ones you should focus on. Now the creative part starts. Traditionally for me FPGA's and Power Supplies have been the biggest cost savers so that's where I start.
FPGAs
Thanks to Moore's Law marching along FPGAs today cost about 1000% less when than FPGA's 10 years ago. Migrating FPGAs to newer families and consolidating multiple FPGAs into a single device is generally the largest cost savings you will find on a card. When approaching FPGA Cost-Reduction I start out by creating block diagrams representing the FPGAs on a card, including cost, IO requirements and densities and see what I can juggle around using newer devices. At the time of writing this article of particular interest are Spartan-3's from Xilinx and Cyclone-2s from Altera but don't forget about Actel and Lattice. Depending on the application they may come in with a lower cost solution. Once I know what's required I'll call up all of the FPGA Suppliers I know (I usually call the supplier who's parts I'm replacing first, depending on volumes they are probably the most motivated to provide a new solution) and let them know what I'm looking for and ask them for proposals. I look at suppliers as partners, without you they have no market and without them you have no product. Building good relationships with suppliers is crucial. Keep in mind that not all FPGA suppliers are created equal, some provide better tools than others. Some provide better support, etc. With better tools and support it might cost less and take less time to develop even if individual device cost is higher. Keep in mind that most FPGAs need to be programmed possible adding extra cost. If no S/W changes are allowed on a CR Design you might have to get creative with programming and might end up with an FPGA that can;t be re-loaded in the field but in an emergency situation S/W can probably be changed to fix an error in an FPGA. Make sure the H/W hooks are in place even if the current software doesn't support them. Finally, merging FPGA code that already exists or re-targeting to a new FPGA takes effort. More effort is required depending on the quality of the original code. Asynchronous code will probably have to be re-written (and probably should have never been written in the first place). If there were no timing constraints they will have to be generated. At the least the code, including test bench should be reviewed in its entirety. You can may also want to take a look at migrating FPGAs to ASICs, but for anything expect for the highest volume designs this is generally cost-prohibitive.
Power Supplies
The second biggest cost savings is usually found with power-supplies. There have been a handful of new lower-cost isolated-module manufactures enter the market in the last few years. It's usually just a matter of shopping around. Keep in mind though that not all power supplies are created equal, make sure you test any new modules before dropping them in a design. Non-isolated DC/DC converters also provide some cost-reduction potential. Usually you can size-down the supplies on a card, migrating to newer technology ICs usually has the added bonus of requiring less power. You can also convert modules into discrete designs, TI's SWIFT family of parts are easy to use and usually provide a cost reduction. For low power converters take a look at MC34063's. They have a tone of different suppliers and are pretty cheap. Using them you can design a 200mA converter for less than $1.50.
Other Devices
Finally there is everything else, Processors, Memory, ASICS, etc. Processors may be expensive to change, due to S/W development costs but you might find a lower cost version in the same family that lacks features that aren't used. You may also find the same device in a different package that represents a cost savings, talk to your suppliers. For memory, moving to whatever the PC Industry is using can represent a cost savings but changing memory will almost always require a S/W or FPGA change and may not be worth it. With ASICs and other devices you may find that moving functions into FPGAs can reduce cost (yes depending on volumes moving an ASIC into an FPGA can sometimes reduce cost). You may also find vendors with newer, lower-cost solutions for "other devices" like newer T1 Framers, clocking devices, etc. I can't cover every possible device in this article. These are just suggested starting points. It's up to the individual CR Designer to get their creative juices flowing and come up with ideas that fit within the framework of the product.
Non-BOM Costs
In addition to BOM savings there are usually other areas of the design that can be changed to cut costs. These are generally more difficult to track before finishing a design, for example a CM will usually need design files before providing a quote. But if you keep these areas in mind while working on the proposal and design you can maximum your employer's savings. Usually I've found that I can find some savings in the PCB itself. I usually use the rather arbitrary rule-of-thumb estimate that each PCB feature costs 10% more, for instance an extra two layers, 10mil vias (vs. standard 12mil), buried capacitance, etc. If you can remove these extra features you can save cost. I've yet to find a PCB that I can't hack layers out of, one card I reduce from 18 layers to 8. One trick that can possibly save layers is to use 10mil vias under 1mm pitch BGAs and run two 3.5mil tracks side-by-side between the vias. If you can reduce the layer count by 4 then it more than makes up for the extra cost of the 10mil vias. I tend to look at buried vias and micro-vias as a luxury. I've never been in a situation where I've needed to use either, and dumping them will make the card easier to layout, test and debug. Unless you can make use of the technique described above 10mil vias are only really required if you have BGAs with 0.8mm pitch or less so I usually avoid these devices. Buried Capacitance is a feature that you can usually do without as well. Though it's virtually impossible to confirm the effects of removing it from a design via simulation so removing it is a bit of a judgment call.
Manufacturing costs can be minimized by maximizing yields. It will take a conversation with your manufacturer to find out what they prefer but usually 0603's have higher yields that 0402s. If using 0402s balanced land-entry may provide higher yields. Following component spacing rules will result in higher yields. Reducing the number of process steps will result in lower costs, for instance if you use only surface mount components on a single side it will only take one pass through the oven and represents minimum cost. Finally, you can look at yield data on the current card and change components or layout features that are reducing yield.
Manufacturing Test Costs can represent a significant portion of Manufacturing Cost. If you can reduce the amount of time every card takes to be tested you can potentially incur significant savings. It may require adding some extra hardware that may or may not pay for itself in reduced test costs or it may be as simple as adding more test points. Remember, the earlier a defect is caught the cheaper it is to fix. If possible I usually try to design cards that test themselves and only require external loop-back connectors. Have a conversation with the Manufacturing Test Group to see if they can make any suggestions to reduce cost.
Finally I would like to mention debug feature costs. I usually don't add extra components to a production card like LEDs for debug because they can add unnecessary cost. If they're in the Alpha build I will usually depopulate them before production or just leave them off and add test-points that can be connected to external hardware, like a logic probe.
Other Miscellaneous Advice
Below is some random additional advice that may or may not be useful. One strategy which has proven successful is to use a CR Re-Design to put pressure on a supplier of an OEM Module. They of course add markup but they may have already amortized development costs and may get volume discounts that aren't available to you so it's not necessarily a slam-dunk. Generally when the OEM finds out that you are working on replacing their product with an in-house design they lower they prices to the point where the internal project is canceled. Which isn't a bad the thing, the ultimate goal of cost reduction was achieved.
I mentioned it before but thought I would mention it again for emphasis. CR Re-Design provides a great opportunity for quality improvement. Make sure you take the time to look at statistics for product returns and manufacturing yields to see if there is any room for improvement. In the end this will also reduce cost.
Another advantage of CR Re-Design is less obvious. Generally CR Projects take less time than a new design (usually 6-12 months compared to 12-24) so a smaller design group is required to CR the same number of products that a New Product Development Group can design from scratch. If there are fewer designers churning through the designs in your company you will end up with more uniform designs, more common components, etc. Which will have a positive impact on specific component volumes and total Company wide component costs. These gains can also be achieved by a pro-active Component Engineering Group but due to the inevitable political issues that route tends to be less effective (unless given the full support of upper management).
Finally, in my experience having designers that design new products work on a couple of Cost-Reduction re-designs will give them a new appreciation for cost. Cost should be a factor in any design decision, no matter the volumes. Cost relates directly to profit, share prices, bonuses. It's an area that an Engineer can have a significant impact on the bottom line. By always keeping an eye on cost an Engineer can help ensure the long term viability of their employer, and ultimately their own career.