PCB Trace Corner Arc Radius

Corners aren't an issue until several GHz. Vias are invisible at 100 MHz too. You only need to control trace impedances for traces longer than, say, 1 foot or so at 100 MHz.

But a clock can have fast edges regardless of frequency, and you don't want ringing or flat spots, so impedance control and termination are a good idea. Crosstalk can be a killer on clock lines, too, so keep other signals away, or on the other side of a ground plane.

We often add tiny-logic schmitt triggers at devices like FPGAs to ensure clean clocking off a bussed clock. Some devices are very sensitive to clock quality.

John

Thanks John That's very germane advice on clocks, I did schmitt it, but quite a way from the FPGA - I'll move it. Gary

We've had trouble with the CCLK (serial configuration clock) input on Xilinx chips. It's super-fast and very tender, so we usually schmitt it right at the chip. I hear that the new chips will start to include schmitts on CCLK.

It's a bitch: if the edge is too fast, it can ring and double-clock. If too slow, truly tiny amounts of noise on either edge can also multi-clock it. There's few things more annoying than an fpga that configures intermittently.

John

I've seen that issue at clients sometimes. Was called out for EMI problems, found some of this, put in a weak AC termination and then it stopped splattering into the 2.45GHz links and such. Miraculously some other glitches disappeared where everyone thought they were software bugs. AC termination sure is nice but for some reason I often have a hard time convincing people to put it in.

Other times I placed an RF transistor as a follower but that requires the presence of some negative voltage to work. It helps reducing the load capacitance at the end of the trace.

Got to Schmitt those inputs ;-)