Custom chips can offer reduced power consumption, compared with a collection of standard parts. Paul Double reports.
A common misconception is that custom ASICs cost a fortune in non-recurring engineering (NRE) charges and they take many months, if not years, to develop. While this may be true for some digital devices, particularly those based on deep sub-micron processes, it is not the case for analogue and mixed signal chips, where even products that will be manufactured in modest quantities can benefit from the approach.
Remember, most mixed-signal chips are produced using mature processes, which means using fully depreciated fabs with much lower costs than parts that demand leading-edge geometries. EDA tools are cheaper than their heavyweight digital counterparts too.
Why design a custom ASIC?
The most common reason for designing a custom integrated circuit in a product is to reduce its cost. In high volume, a custom IC that replaces several standard parts is often less expensive. However, even for lower-volume products, custom chips offer many other advantages. For some applications, high performance is paramount. No matter what the typical gate delay or clock frequency, it can be improved by a factor of 10 with a custom chip. Try to architect the system so that only a small portion of it requires maximum-speed tweaking. The rest can be created by automatic design software at more relaxed performance requirements.
A custom chip can offer reduced power consumption, compared with a collection of standard parts. The energy required to drive a node to a high or low voltage depends on the capacitive and inductive load, and these loads are smaller for internal nodes. So, if many of the external nodes of the PCB design can be changed into internal nodes of the custom IC, it will reduce power consumption. Furthermore, there are other energy-saving techniques that are unique to custom ICs.
A designer with access to the full flexibility of a custom chip can create numerous special functions that are difficult to find elsewhere. For example, special purpose arithmetic units, multi-port memories, and a variety of non-volatile storage circuits can be developed. One can even create magnetic sensors and light sensors ranging from a single sensor to line sensors and two-dimensional video camera chips.
Some companies use custom ICs to better protect their intellectual property. A custom integrated circuit is much more difficult to reverse engineer than a board level design.
Higher integration levels bring greater system reliability. If the board has dozens of parts and hundreds of solder connections, and it can be replaced with one or a few parts, with fewer board-level interconnects, the system becomes more reliable.
Likewise, higher integration leads to lower manufacturing costs. If the custom-IC solution uses less power, a cheaper power supply can be used. Fewer boards also mean fewer connectors and smaller, less-expensive cabinets.
One company built a product that had two discrete transistors, a photocell, and a few resistors and capacitors. The circuit board was larger than they needed and they had a measurable field failure rate and a cost of about $1. The company designed a custom IC with several thousand transistors to implement the same function. It had no measurable field failure rate, and the unit cost was about $0.50. For the millions of units sold, the payback on this custom chip investment was rapid.
Custom chips have higher tooling costs, so if it's important to minimise the cost of prototypes, use standard parts. With a few printed circuit boards and a handful of parts, hand-soldered together, a prototype can cost about $2000. The tooling costs of a custom IC start at about $18 000 for a set of masks for a 0.6µm process and go up to about $3 million for a 65 nm process. Products that have high volumes and require huge amounts of processing and memory will need the finest line width processes to get the lowest cost in production.
However, for most other products, the manufacturing volumes never make sense for the $3 million tooling cost. Fortunately, the tooling for coarser line widths is much more affordable, yet still larger than that of a PCB.
Custom chips also have longer lead times. A board can be manufactured in a couple days, and with shipping and assembly time, a new prototype can be built using standard parts in less than a week. The custom route will take weeks, if not months, before the first chips arrive at your door. And although expediting is often available, the fees are steep and shave only a few days off a lengthy process.
Similarly if there are many design revisions, printed circuit boards have the advantage. The cost and delay of revising custom ICs could be prohibitive. The best way to mitigate the risk of revising a custom chip is to get it right the first time.
With modern Electronic Design Automation (EDA) software created specifically for custom integrated circuit design, most common errors can be eliminated before incurring the cost of tooling or manufacturing. Debug is performed through simulation, and design rule violations are caught before the design goes to fabrication.
Historically, the cost of IC design tools has been dramatically higher than those for printed circuit boards.
A reasonable PCB package, running on inexpensive Windows computers, can be purchased for $10000. High-end IC design tools can cost $500 000 or more and run on Unix.
Fortunately, there are now IC design suites - such as Tanner EDA's integrated IC tool suite - that cost about $25 000 and run on inexpensive Windows PCs.
The first custom chip design should not be a high-pressure, bet-the-company kind of design push, but on the other hand, if there is not enough pressure to do it, it probably won't get done.
So look for the right project and consider how such a project may benefit from some of the advantages described here. If the advantages outweigh the disadvantages, jump into custom integrated circuit design.
Paul Double is Managing Director of EDA Solutions in Southampton, UK. www.eda-solutions.com or www.tannereda.com.