Discontinued chips need not be a problem
A good contingency plan can soften the impact of end-of-life notices for key semiconductors in your design, writes Dan Florence, Product Marketing Engineer at ON Semiconductor.
Many companies who manufacture electronics require contingency plans for their products. If a memory device is discontinued, we switch to a competing device. If our circuit board supplier goes bankrupt, we have an alternate vendor on speed dial. But some portions of a system are more easily replaced than others.
Custom logic devices such as FPGAs and application-specific integrated circuits (Asics) can be particularly difficult to substitute.
End-of-life notices rarely come at convenient times. In industries with longer product lifecycles, it is not unheard of for a semiconductor component to be discontinued before a product even ramps to full production. And a last-time-buy is often not feasible; you are forced to accurately estimate your future lifetime usage of the chip, buy that entire amount at an inflated price, and pray you won’t have issues with shelf life or product quality.
The cost of replacing an essential element in your system design can be high. Board re-spins are typical. Your Asic may use a proprietary package, which can make replacement difficult and require a custom package development.
If your custom logic device is an FPGA, the next-generation chip will often have a different core voltage, pad layout and package footprint, and will therefore not be drop-in compatible. All these differences mean you will have to re-design your board to fit your new device and account for routing, voltage and timing differences.
Depending on your quality procedures, these changes may require a costly product requalification. In military and aerospace applications, this can mean millions of dollars. At the very least, system designers will need to spend significant time and energy integrating the replacement microchip into the system, re-characterizing based on old design files, and testing the modified system thoroughly to ensure it is production-ready.
While the scrambling is going on, reduced production or even full line-down situations are not uncommon. What is the cost of reducing your production by half for a few months? Or shutting it down completely?
And it’s not just the immediate hit of lost sales – how will your company’s image and reputation be damaged by an inability to ship product?
The good news is that you can begin protecting yourself against EOLs and other major component disruptions long before your system is even designed. Proper vendor selection can help mitigate your risk, and sound design and archiving principles can make it so your contingency plan is actually a plan instead of a baseless wish. And if you’re already in a bind, you may still have options.
Understanding your vendors’ business models
Product EOLs are often a pure business decision. Utilization of a fab line eventually drops over time, and the device you’re purchasing becomes less profitable to the manufacturer.
All semiconductor companies have finite financial and engineering resources; when they’re trying to launch a new technology or strengthen their mainline offering, it can become very difficult to support legacy products. At some point it just makes more business sense to close down a line than to keep it running.
You need to understand your vendor’s long-term strategy before you even issue that first purchase order. Where are the focusing their efforts? What was the last product for which they issued an EOL? Where is the device you’re looking to purchase in the supplier’s product lifecycle?
If the vendor is fabless, i.e. they don’t own manufacturing facilities, but subcontract to foundries, the situation becomes more complicated. If the supplier has deep relationship with their foundry and is a high-volume wafer purchaser, they may have some sway in negotiating when a product is discontinued. But even then, the ultimate decision to close a line is out of your vendor’s hands.
This doesn’t mean you need a cutting-edge technology, and its accompanying price premium, to ensure process longevity. Every semiconductor manufacturer takes a slightly different approach to lifecycle management, and some business models invest more heavily in mid-range or even legacy technologies.
In particular, digital processes that serve as a backbone for analog and mixed-signal development are likely to be in production for a long time. Because analog development trails digital design by a few technology nodes, analog semiconductor companies often keep their digital processes running much longer than digital-only vendors.
Ask potential custom logic suppliers about their business model. Ask around your industry about their process longevity. Search for EOL notices from them and try to get an understanding of how long their processes are typically supported. A little research at the beginning of a project can save you from major headaches in a few years.
Selecting industry-standard rather than proprietary IP
FPGA and Asic providers sometimes offer free or heavily discounted intellectual property, which can be very helpful in the early stages of the design process. This helps you, the developer, have a very low start up cost for a project, which is ideal when you are vetting a new product concept or strapped for R&D financing. But if the IP is proprietary, you are committing yourself to the vendor’s platform.
It’s best to select industry-standard IP, which is built conform to a universal specification, when creating a custom logic device. Should you later need to change processes or vendors, this IP is largely portable – you can often purchase a license directly from an IP provider, or at the very least buy a compatible offering.
A DDR2 interface, for example, is basically always a DDR2 interface, and although some minor changes may be required when moving from one vendor to another, the rest of your design will remain largely unchanged.
On the other hand, custom IP blocks, especially microprocessors, are often very difficult to transfer. Even if it is technically possible to migrate IP from one vendor’s process to another, the original vendor will either not allow the IP to be licensed for use elsewhere, or will allow it but only on cost-prohibitive terms. You will have much less dependence on a particular manufacturer if you build your design using portable, standard IP.
Properly archive design files
Let’s face it: design files are complex, large, and sometimes scattered between individual contributors. They are often stored in various states of revision and development branches. If your data is a little confusing now, just wait until you’re trying to sort everything out a few years from now.
If you ever need to migrate a custom logic device, one of the first things you will need is your original design data, and it will need to be complete. RTL or netlists need to be saved, but you will also want timing constraints files, synthesis scripts, pin lists, block diagrams, IP information, memory configurations, and clocking structure. Simulation test benches and board-level information can also aid in quickly and accurately converting your design to another semiconductor process.
Clearly label and store your data in a method that will be accessible and locatable years down the road. Review and perhaps update your data storage procedures to account for possible retention needs far into the future. Sloppiness or laziness in the design phase could cost your company hundreds of thousands or even millions of dollars in future conversion costs or lost revenue.
I’ve received an EOL already – help!
If you find yourself in the unlucky situation of having already received a discontinuation notice, now is the time to begin examining your options. Of course you will reevaluate your product forecasts and determine how large your component gap may be, but you should also talk to new suppliers as soon as possible.
If you determine a migration would be beneficial, ideally you will be able to get samples or even production volumes of your new device before the last time buy of your current device. However, logic migration takes time and resources, and the more you delay, the more your options will narrow.
Some custom logic manufacturers specialize in the conversion of existing products. Still others may have process lifecycles that will be sufficient for your long-term needs. Vendor selection for a replacement device will be just as critical as it was for your original design, if not more so. Make an effort to understand your options and find one that aligns with your plans and requirements.
At the very least you will identify ways to avoid EOL issues for your company in the future. Take the time to document and share those lessons learned, and help institutionalize procedures that will minimize discontinuation risk. As you will know, end-of-life events are worth avoiding.Asics, chips, design, FPGAs, ON semi