SMDs and 4-layer PCBs

I used to do freelance PCB design so was pretty well into it (anyone remember Bishop graphics and DOS Protel?), but I'm currently doing a board after quite some time off (maybe 20 years) using SMDs and 4 layers and it seems the rules have changed a bit.

In the past you would religiously run one side vertical and the other side horizontal. The second you snuk a horizontal trace vertically (or vv) by even a few 10ths you would cut your water off by blocking a future trace. At the end you could rip up and re-route when you had no more traces to run but that rule was absolute until then.

Now it seems all different. Many SMDs don't allow traces between pins so they often totally block a large area of PCB on one side. For example a the SAM3U, 144 pins all of which have to be broken out with no possibility of routing between them. So regardless of where they are going they all have to start on the component side.

In other places there are components on both sides so all bets are off and you have to be careful not to really box yourself in.

Now I find myself just running traces on whatever side gets them through and to hell with the rules.

On the plus side is 4 layers. What a breath of fresh air to not have to find space for large power traces. Just a short stub off a pin and a via to a middle layer. Great stuff.

Anyway just a couple of thoughts about the current state of the art as I see it.

Is it fair to say that the old horizontal/vertical rule no longer applies (or at least not as much)?


Rob

Some people still do their boards that way, even surface-mount ones.

The usual routing strategy these days is to route the critical tracks like supplies, decoupling and clocks.Then route the shortest tracks, and so on, until the whole board is routed.

Yeah that's basically what I'm doing, it's easy to add functions (and therefore ICs) when doing the component layout but the package density can trip you up, all those signals have to get out and there are so many no-go zones. Without 4 layers I wouldn't have a chance and the board would probably be twice the size. Admittedly once I decided to use one of the internal layers for signals I got lazy and stopped looking for routes on the top and bottom layers.

So far so good...

Only a few traces to go plus add the ground plane.


Rob

Is it fair to say that the old horizontal/vertical rule no longer applies (or at least not as much)?

I think it was only a useful rule when you were laying out a PCB with tape, because you needed to minimise the number of times you changed the tracks. Now with computers it is easy to move tracks. Anyway boards laid out like that never looked very nice in my opinion.

I used to stick rigidly to the H/V rule, even if it meant two vias and a .2" trace because that tiny gap was enough to sneak a trace through and get you out of trouble. Towards the end of the layout when you where 95% sure you wouldn’t need to run (m)any more traces in a given area you could get more creative and not be so pedantic about it.

Once the board was totally laid out I would spend another day or so removing as many silly layer changes and vias as possible. But as you say it was difficult to make mods with tape so you were less included to do so. It was also a good way to add bugs if you ripped up a few pieces of tape then went to lunch :slight_smile:

At least now we get a flying wire as soon as we delete a trace, not to mention the DRC.


Rob

All done, still have ground and heatsink planes to add but this is the almost-finished result

Man I love having those extra layers, I remember how great it was going from 1 to 2 layers all those years ago and this is a similar revelation, with 4 layers and SMDs on both sides you really can pack 10 pounds of sh1t into a 5-pound bag.

Now I just have to convince 'er indoors to let me get it fabbed :)


Rob

The last project I worked on for a payed job, had a 12 layer board. Even then the PCB guy struggled to get all the lines out from a 1000+ pin BGA chip. The board was big as well over 400mm by 300mm

Smd definetly adds some challenge for example its 10 easier to route and fit components with a 328 dip, than a smd version in the dip I can pit the crystal,caps, and a few resistors under the socket, and never have a problem runninft traces, with tthe smd and the same components take up. More space and I usually cramn it all and make it fit then run traces , alot harder with smd

12 layer board...1000+ pin BGA chip...400mm by 300mm

Unbelievable, makes my little board look pretty simple.

I guess one "problem" with SMDs is that you are tempted to pack lots in because the chips fit, but there's still the same number (and often a lot more) of traces to run and a lot of the board is out of bounds because you can't run traces between pins.

For example a 28-pin dip is ~1.3" long and in that space you can run about 26 traces between pins. The TSSOP28 is only about .35" long but you can't run a single trace through it.

Net result, less real estate for traces if you pack the chips in.

So as I see it you have to start leaving more space around chips or go to four layers.

After doing this board four layers definitely gets my vote.

I can pit the crystal,caps, and a few resistors under the socket,

You can of course do this with SMD as well, just put them on the other side of the board. I did a small PCB a while back with a tiny85, rs485 transceiver and a few small components, all up about 1/2" square and that's using SIOCs that are relative large by SMD standards.


Rob

I usally have the crystal on the back, unfortunetly its not smd and I don't like cutting the pins flush on the other side to fit the ic flush still another thing about most smd is that the fets and transistors are meant to lay horizontal and take up more space than the same dip that can go vertical, I've seen it done on my old cars ecu actually they had some smd components on a smaller pcb that was soldered perpendicular to the main board just to fit more, I think I may copy that technique for a board or too I have in mind

Yes good point, PTH power devices often take up less space because they can be vertical and don't need a large heatsink plane around them as well. Bottom line is that power devices pretty much have to be a certain size for a given power and technology, you choose if that size is horizontal or vertical and in many applications vertical is using enclosure space that otherwise is not used and in fact you may even get a "free" heatsink in the metal enclosure.

In this case even a large PTH power transistor takes up almost no PCB space whereas it's SMD equivalent uses half the board.


Rob

One thing I like about smd is how nice it looks when everything is smd, very sleek and professional im actually thinking about getting a sleeve tatoo one day with a self designed functional board , thinking about having part of it so functional that I could trace it with that silver ink trace pen and put a tqfp 328p with some smd leds, maybe a small lipo battery taped to the underside of my arm for power and have the coolest single layer board out there

12 layer board...1000+ pin BGA chip...400mm by 300mm Unbelievable,

This is a publicity shot of me "examining" the board. http://www.flickr.com/photos/33177304@N03/6305319953/ You can see there are sockets for 10 extra boards to be plugged in. In all there were just over 8000 components in the system. .There are five large FPGAs on the board

Wow, what could that be used for? I imagine a single board must cost thousands

Well close, the full unit was about $28,000 (boxed) and it was a very configurable transmodulator, taking satellite digital signals in at one end. Stripping the multiplex down into its component parts, assembling a new multiplex from those parts and re-modulating it as a digital cable signal. A sort of mini head end. It also allowed high speed data connections between the cable end and the box thus acting as an internet gate way, and multi-channel satellite system over poor quality coax lines that were already installed in buildings as communal antenna systems.

it was a very configurable transmodulator, taking satellite digital signals in at one end. Stripping the multiplex down into its component parts, assembling a new multiplex from those parts and re-modulating it as a digital cable signal. A sort of mini head end. It also allowed high speed data connections between the cable end and the box thus acting as an internet gate way, and multi-channel satellite system over poor quality coax lines that were already installed in buildings as communal antenna systems.

You know that was my first though when looked at the photo :slight_smile:

BTW I love those head magnifiers gadgets, the combination of older eyes and smaller components is not good but they save the day. I’ve got one with lights that illuminate the subject as well.


Rob

I think I got that same visor, I love it!

The H/V rule still applies for a lot of people. I try to use it but as my main job in electronics is prototypes anything goes. Once I have a working prototype I normally email it to mmcpix and he does the business with the production board. Big differences between prototypes and production re things like acid traps, track angles and the like. I also like the have all components soldering on the right side rather than component legs acting as vias (for prototypes). Just to give you an idea, mmcpix just traced a board for me with: SMD 1284p-au; SMD resonator; 2x3 header; 1x5 header; 1x6 header; 2 x SMD caps; 3 x SMD resistors; battery; 8 pin DIL and a 28 pin DIL header in 50mm x 50mm on 2 layers. :)

I can do them myself but he always complains that they are messy or break rules :) His boards do look neater though.