On 6/25/07, Jens M. Guessregen / Mailinglists wrote: > Ok ... > > > > By the way, high speed differental and impendance > > controlled busses on > > > the outer layers is a bad idea, just making the layer stack very > > > inflexible and expensive. > > > > Doing it on the inner layer as stripline instead of > > microstrip on the outer layers will be the same as long will > > be used only through holes vias and back drilling is a > > complicated issue. A through vias on a 3Gbps line means an > > unwanted stub between the used line and the bottom layer. > > Well, first of all, you should make clear to your self, how a 12 layer > pcb is build up. > Typical, a 12 layer is build out of 2x 6 layer packages. A 16 layer is > build out of 2x 8 layers. You can also buil 14 layers like that from a 6 > and a 8 layer, but that asymetrical design can cause issue under the > vakuum pressure, the PCB has to go through. On a previous design I've made three separately designs (stacks, routing etc) untill PCB house was able to produce it. I promised to myself I'll never design boards other than with simmetrical stack, and never before the PCB houses gave me complete informations. > > So, at least, you have to packaes out of a 6 layer. Take the inner 2 > planes as main power supply, use layer 2 and 5 as impendance controlles > layers and take the layer 1 (top layer) and layer 6 for fanout and easy > routing (means stuff without high requirements). The same for the second > package. This sound interesting. Is not like any standardised 12 layer stack I know (from 12A to 12L standard names). Correct me if I understood wrong. Your stack proposal is: ------------ start sublamination 1 1-top 2-signal 3-PWR 4-GND 5-signal 6-signal ------------ end sublamination 1 ------------ start sublamination 2 7-signal 8-signal 9-PWR 10-GND 11-signal 12- bottom ------------ end sublamination 2 What will be hapenning between layer 6 and layer 7 ? There will be a great EMI there even layers 5,6,7 and 8 will be routed using 45 degree and ortogonal schemes with as much ground copper as possible. Will be only three types of through vias (through vias per sublamination, through and blind vias per stack) 1-6, 7-12, 1-12. > Regarding this, you go through a via with your 3 GPS line from top layer > (assembly layer) to layer 5 (routing layer). Having 0.8 mm as thickness > of the 6 layer stack, is leaving you a stub of ~0.1 mm (from the last > pre-preg). I think, that is very acceptable. > > In the final stack with 2x 6 layers, that is looking like a blind via, > but it is produced like a standart via, so extra costs for that should > be minimal to nothing. > > If you can not go with the ,1 mm stub, you should have a closer look to > plugged vias or partitial plugged vias. With that, you can have standard > via's (not microvia's) inside landing pads without getting trouble, when > soldering is disappearing into the via's. This is offered f.e. by ILFA > in Hannover/Germany. I've guess is this one right ? http://www.ilfa.de/servicecenter_en/ Are you with ILFA ar an ILFA customer ? I'll try with my next design, I'm curious about prices. For example how much cost a 12 layer board as follows: 2500 routes, 2200 microvias/buried vias/through hole vias, min drill size 0.127mm (5mil), min route/isolation thickness 0.1mm (4 mil), overal board thickness 1mm (~40mil), size 140x50mm ? > Costs are small, even more, if you can reduce layer > count at the same time. Plugging is sparing a lot of space on the > toplayer, giving you space for direct routing, if this is prefered, but > then, you have to look very close to impendance. You will need a GND or > VCC plane as 2nd plane. And regard, that copper thickness on outer and > inner planes is different, having influence on impendance and other > issues also. Yes, I think I have no problems with propagation of waves in free space (stripline) and in controlled dielectric space (microstrip or coupled microstrip). I'm computing every critical routes. > > > > You should start with a clear stack setup for your impendance, then > > > fanout all power and GND and then routing only the impendance > > > controlled lines. After that, you can see, if you need additional > > > layers for the rest or not. > > > > Yes, but that starategy implies a very difficult stack-up > > modification. > > Nobody said, that 1100+ BGA is easy to go with ;-) > But my experience is, that you are not able to calculate everything > before having anything placed and routed. You need to spend time to go > after try&error ... > > My typical way is that: > > - Place parts by hand with knowledge of function and important lines, a > perfect placement is half of the job > - if needed, think about changing pinout of FPGA > - Fanout for VCC and GND > -Routing important lines by hand > -Make Backup of this state > -Try out of autorouting all "normal" and "static" lines with a good > autorouter, just to get an idea if there is a close chance to get it > routed with the selected stack > - If the autorouter is coming up to <10% open lines left, a good manual > layouter will do the job > - Delete this autoroute tryout (or store it somewhere else for > reference) > - reload Backup and do layout by hand with interactiv manual routing > - If the autorouter is not able to get closer as 10%, you have to rework > layer stack setup, but at that point, it still is easy, using the right > tool. > - do not use the autorouter to do the complete job. You have to spend > more time to clean up, as to to do it by hand (interactive) from start. Yes, I see I'm using more or less the same tehnique. But remember 3% unrouted could be 50...100 routes... :( and 100 routes for manually routing *after* the rest of the routes have been routed, could be one-two weeks of hard work. I care to my neurons and eyes. thanks for suggestion, Vasile > > > > I don't know what cad is your favourite, mine's > > is the one company bought. > > Yes, but that must not be the perfect one for that job ;-) > > I am working with Mentor Expedition, since 98, when it still was > Veribest ... > I worked also with OrCad, Protel and Eagle, but at least OrCad and Eagle > are looking like playtools against Expedition. > I do not like the handling of Protel, but that is more my personal > experience. > > Best Jens > > -- > http://www.piclist.com PIC/SX FAQ & list archive > View/change your membership options at > http://mailman.mit.edu/mailman/listinfo/piclist > -- http://www.piclist.com PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist