Multiple PCB system vs mixed-signal PCB

Question

I'm designing a high-frequency (\$6\,\text{GHz}\$) mixed-signal PCB. The design contains some noise-sensitive RF analog circuitry (mixer, LNA, 12-bit ADC, etc.) as well as some noisy digital circuitry (e.g. FPGA driven by a \$40\,\text{MHz}\$ clock with \$1\,\text{ns}\$ edges) and several switching converters. The design is small enough (a bit less than 10x10cm) that I'd previously assumed it would be a single PCB design (admittedly, I've never designed a system with more than 1 PCB). Despite the size, I'm beginning to think that this could be better suited to a multi-PCB design. The first benefit is being able to separate the analog and digital circuitry. I can individually shield these PCBs and I expect should be able to reduce the digital noise that creeps into the analog system.

Another benefit of separate PCBs is it makes the system more modular and easier to test. In particular, my current thought is to have 3 separate PCBs for the power supply, digital circuitry and analog circuitry. This would allow me to more easily test the power supply and substitute in different power supplies if I think one is too noisy or sub-optimal in some other way. Or, I could generate dummy data for the digital section, or command the analog section externally. You get the idea.

The final reason I'm intrigued by the multi-PCB system is cost. The RF analog portion requires a significantly more expensive PCB than the rest of the design. I can place the power supply and digital PCBs on cheaper (higher loss) materials and save a substantial amount without (I hope) sacrificing quality. I think the digital section is slow enough that the higher loss and less controlled dielectric permittivity shouldn't matter. Moreover, if I mess up one of the boards the cost of replacing just that portion is lower.

However, there are a number of challenges with the new design. For one, I run the FPGA, the ADC and an RF frequency synthesizer on the same fast edge 40MHz clock. The FPGA would be placed on the digital board whereas the frequency synthesizer and ADC would go on the analog board. If I place the clock on the digital PCB (which was my plan) I would need to ensure the clock signal and its return current have a low inductance path between the digital and analog boards. The same goes for the digital output of the ADC but in the reverse direction. Similarly, there are a few digital control signals that are sent by the digital board to control the analog one. I believe the solution to this is to use the proper board-to-board connectors. I'm not too familiar with these, but considering PCIe exists, there must be a number of varieties that are more than adequate for my use case.

Could this design benefit from being on multiple PCBs for the reasons I provided? Are there other reasons I haven't considered that are either for or against doing this? Any thoughts on the proper board-to-board connector to use for the clock and ADC digital data? All of my voltage and current levels are modest: the full design takes 12V 600mA.

I'm happy to provide more information about this design if it's useful, just let me know.

Answer 1

There are (as always) pros and cons; much depends on just how much more expensive the laminate is for the RF section (I know of a laminate that can handle 6GHz quite well that has about a 25% premium over more standard material). It will also depend on the complexity of the layout(s) involved; using a separate board for the RF section might well reduce that.

There are one time costs (assuming you are going to make less than a few hundred - some things do need to be replaced).

Truly one time is PCB tooling (assuming you do not change the design - if so there will be a tooling cost for each iteration).

Solder stencils are good for at least a few hundred (and quite possibly more) runs on an automated line.

A word of caution; the largest source of problems in PCBs is (in my experience anyway) is connectors but that depends on the number of mating cycles - if you mate up connectors (and have a positive locking mechanism) then you may not run into that problem.

Could you benefit from multiple PCBs? Perhaps - only by evaluating the options (and talking to PCB houses about costs) will you actually know.

To the details.

Clock distribution; you could use a clock buffer (typical solution linked - there are a number of vendors in this space).

Connectors: My goto here is Samtec as they have a really great range of connectors in the high speed / signal integrity area (there are, of course, other manufacturers). I have successfully used their connectors in many high speed / high throughput systems; as you believe, there are definitely connectors available (from a number of manufacturers) that will meet your needs.

My personal opinion (and it is only that) would be to go for 2 boards (digital and power on one). It is perfectly possible to do a single board, but that might require more effort in design (keeping digital noise out of a very high speed analog signal is a bit of an art form). I went into that subject (as indeed have many others on this site) here.

There are also a number of excellent resources available online.

Make sure that you have a very solid low impedance ground for all the sections (so that the return currents are all referenced to the same potential) and do any filtering to the RF section on the power side (either with a very low noise regulator or a ferrite filter (The details of what to use are very application dependent).

So you might benefit from a multiple board solution (certainly it will help you separate the analog section from the digital noise although you will need to take care with the ADC routing).