## Submicron ASICs and EMI/EMC

Kenneth Webb writes:

Do you have any references (or thoughts) pertaining to emissions and susceptibility of equipment when changing an ASIC from a 1-μm process to a 0.5-μm process? The "new" ASIC is a form, fit, and function replacement for the original, and it comes in the same package style.

Let's first talk a little about the general form of the spectral-power-density for a random digital signal. In this discussion, you define the knee frequency, F_{k}, for any digital signal as F_{k}=0.5/T_{r}. The variable T_{r} represents the signal's 10 to 90% rise or fall time, whichever is less.

The spectrum of a random digital signal has three regions. First, from dc to F_{k}, the spectral power density generally falls off with a slope of -6 dB/octave. If the signal is a repetitive clock, it has a spectral line structure, but the peaks still fall off at a roughly -6-dB/octave slope. Second, in the region near F_{k}, the shape of the spectrum foretells the shape of the rising and falling edges. For example, a large resonance near F_{k} indicates overshoot and ringing in the time-domain signal. Lastly, at frequencies greater than F_{k}, the spectrum rapidly falls off. For example, at two to three times F_{k}, the spectral power density may be 100 dB less than the level at the data rate.

To calculate the intensity of circuit emissions at any given frequency, *f*, you
need to know the antenna effectiveness of the radiating circuit. As a general rule, the antenna effectiveness of small objects, such as chips and circuit traces varies in proportion to *f*. The rate of increase is roughly 6 dB/octave. Use that raw assumption for now.

You calculate the overall emissions at frequency *f* by multiplying the spectral power density of your digital signal times the antenna effectiveness of the radiating circuit.

In the frequency range from dc to F_{k}, the spectral power density of the data signal decreases at a rate of -6 dB/octave, and the antenna effectiveness increases at 6 dB/octave. Given no other in-formation, you may conclude that the emissions from a random signal should be relatively flat from dc to F_{k}.

At greater than F_{k}, the spectral power density decays much faster than the antenna effectiveness rises, so the overall emissions decrease with F. In this region, the limited rise/fall time in your chips provides a natural filtering effect that limits emissions beyond F_{k}. Of course, in real life, you may encounter huge resonances, harmonic spikes from repetitive clock signals, and so forth, but the general form of the response given here represents a real signal.

Now let's get back to Webb's question. If he halves the rise/fall time of his circuits, F_{k} will increase by a factor of two, shifting the emissions to the right by another octave. For example, if Webb has 1-nsec edges in his circuits, the natural filtering effect of those circuits starts at about 500 MHz. At greater than 500 MHz, he probably has few emission problems. If his new circuits have 1/2-nsec edges, their natural filtering doesn't start until 1 GHz. All his natural protection in the 500-MHz to 1-GHz region evaporates. Kenneth will then have to ensure that his EMI filtering, grounding, and shielding means are effective in the new 500-MHz to 1-GHz band (twice the frequency required before). In particular, he may need to cut the inductance of all grounding lugs and grounding connections by a factor of two. He may need a lot of shorter, fatter connections. Webb will probably also uncover a number of resonances and other problems in the 500-MHz to 1-GHz area that may not have been big issues before but now require attention.

On the other hand, Webb may get lucky. If his old product has some margin, meaning that it happens to radiate in the 500-MHz to 1-GHz range at a level far below the Federal Communications Commission (FCC) radiated-emission limits, that radiation margin may allow his new product to pass unmodified. If you—like Webb—are planning to shrink your ASIC, I suggest you check right away to see how much FCC margin you have in the critical region just above F_{k}.