CircuiTree - January 2009 - (Page 35) Intelligent Design By Lee W. Ritchey Are Some IPC Standards Out of Date? ecently, I received an email from a designer asking me how to document trace width accuracy on a controlled impedance PCB. This company had been using standard notes on fabrication drawings that cite a section of IPC specification IPC-6012 that states trace width may not vary more than ±20 percent. This standard has been widely used by both designers and fabricators as the default against which to build and inspect PCBs. Is it out of date considering the changes in technology over the past two decades and especially the last five years? R used in even the lowest-cost PCs and video games can have rise and fall times as fast as 200 ps. Such rise and fall times were only required in supercomputers just a few years ago. The video games often don’t need such speeds, but the fabrication methods used to produce these ICs result in these speeds. At such fast speeds, the length of signal line that can be degraded if impedance control is not exercised is often less than one-half inch! This means that virtually every signal on PCBs using such ICs requires controlled impedance transmission lines. To have specifications that do not recognize this change The portion in question of IPC-6012 was written in the late 1970s when there was little, if any, need for controlled impedance traces. The primary concern was control over imaging and etching in fabrication shops because this was high on the list of reasons for unreliable PCBs. In the ensuing years, the technology available to fabricate PCBs has improved so dramatically that trace width variation has faded into the background as a source of PCB problems. At the same time, the speed of standard ICs used to manufacture everything from video games to super computers has increased to such a degree that virtually every product, even slot machines, requires impedance control to ensure reliable operation. An example should illustrate how the industry has changed and shed some light on the need for revision of some of the key specifications that the PCB industry relies on to assure adequate quality in the PCBs it manufactures. DDR2 memory ICs currently in the overall requirements of our industry seems somehow wrong. To illustrate, with a typical 50 ohm trace achieved with a 5 mil (127 µm) nominal trace width, specifying a width tolerance of ±20 percent results in an impedance range of 46 to 55 ohms just from trace width variations of +10 percent and minus 8 percent using up nearly all of the allowable tolerance in a well-made, controlled impedance PCB. The total allowable impedance error from all sources including thickness variations, variations in dielectric constant, and trace width variations is only ±10 percent for most controlled impedance PCBs. A far better method for specifying allowable trace width variations might be to refer to the current ability of fabricators to control trace width and specify that using class of width control based on capabilities of fabricators, much as is done with annular ring specifications. Similar changes have taken place in the need to control the makeup of the laminates used to fabricate PCBs. When speeds were slow, the primary concern for PCB laminates was to ensure they tolerated the elevated temperatures of soldering and rework, that they be CAF free, and that they be dimensionally stable. With the speed changes illustrated previously and the widespread use of multigigabit per second differential signaling, losses, uniformity of glass weave, accurate dielectric constant specifications at a variety of frequencies, and consistent glass to resin ratios have emerged as key properties to specify and control. I often tell the students in my high-speed design classes that standards document the past and that those pushing the state of the art will have trouble documenting their designs if they rely only on the standards currently in print. This has never been truer than at the present time with semiconductor technology rapidly changing how design is done. As an example, in 2001, one of my clients built the first terabit router for the Internet. It was a half rack tall, weighed 350 lbs, consumed 7 kW of power, and required 51 PCBs. Last year, another client began shipping the same terabit router function in a 1U high rack mounted case. This unit required only one PCB, weighed 22 lbs, and consumed 500 W. If the standards in place in 2001 were followed, the 2007 product would have been unbuildable. With an industry changing at the pace it has in the last 10 years, and changing even faster at the present time, standards need to be updated rapidly if they are to be of value to the industry that relies on them rather than at the snail’s pace currently being followed. ■ Lee W. Ritchey is currently president of Speeding Edge, a leading training and consulting company specializing in the design of high speed PCBs and systems. He has spent his 40-year career designing high-speed PCBs for supercomputers and high-performance Internet products. Email: leeritchey@earthlink.net circuitree.com • January 2009 35 http://www.circuitree.com
For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.