Issue link: https://iconnect007.uberflip.com/i/1457913
46 DESIGN007 MAGAZINE I MARCH 2022 100 MHz SPI clock, we know that the rule of thumb is that a square wave needs at least five harmonics to accurately reproduce the square wave shape. e quarter wavelength length would then be approximately: 1.6/0.5 = 3.2 inches. Any trace less than about 3.2 inches carrying this clock will have poor efficiency as a radiator. Another common rule of thumb is if you don't have a repetitive clock signal but instead have a fast rise-time signal. e bandwidth of a known rise-time is shown in Equation 4. Equation 4 Where Tr is the rise time of the pulse in nanoseconds and f is the equivalent bandwidth of the signal in GHz. It is well known that a 1 ns rise time is approximately equivalent to a 0.35 GHz (or 350 MHz) signal bandwidth. Given these rules of thumb, we can calculate the critical trace lengths in terms of frequency and rise time of our signals. Normally all our logic edges are quite fast now, faster than they need to be for most clock signals, so I nor- mally use the rise-time equation to figure the required bandwidth and then use that to calcu- late the critical trace lengths to watch out for. is "less than a quarter wavelength" rule of thumb applies to all sorts of analysis situations and is a useful thing to keep in mind during the design, and perhaps more importantly, trou- bleshooting of higher-frequency PCBs 6 . The Fix Now that we understand what the critical frequencies and lengths are we can move on with our simulations and solutions. We have just proven that any copper struc- ture, trace, pour, or ground plane can and will act as a resonant antenna if its length gets to or exceeds a quarter or half wavelength of the excitation frequency. Now, look back to the very first picture of a CPWG layout in Figure 1. Notice anything in those copper pours? Can you see the little vias all along the periphery of the copper pours? I didn't put those there just because I wanted to make life miserable for my PCB fabricator. Actually, those vias stitch the grounds together and effectively provide a way to shorten the length of the copper pours to be less than a quarter wavelength. How can we fix Figure 4 then? How about effectively making the copper pour in Figure 4 smaller in any direction less than a quarter wavelength? If we place ground stitching vias in the copper pours we can achieve this, as shown in Figure 7. e addition of properly spaced stitching vias in the copper pour shows that at any fre- quency, now, the central copper pour acts like a ground and never gets hot as it did in Figures 5 and 6, even at the highest frequency of this sweep (Figure 8). When properly done, copper pour is indeed acting as the true shielding ground plane that we had envisioned it would be when we put Figure 7: To fix that hot central copper pour we can apply stitching vias to the ground return layer and effectively make the copper pours appear electrically shorter at high frequencies and make them appear more like the ground they were supposed to be in the first place. Here I placed the vias (dark red squares) symmetrically at a distance of 0.5 inches to divide the central copper pour into sections less than the quarter wavelength at 3.2 GHz.