6
\$\begingroup\$

Manhattan-Style routing being the use of expressly east-west planes and north-south planes, using a via and changing planes when a signal changes direction.

Comparing to freestyle routing, which lets define as routing signals in any direction on a given layer, would manhattan routing generally result in increased density, signal integrity, and more or less layers?

I know this is somewhat general and highly specific to a given application, but I'm generally interested in why one would decide to route in a manhattan-style -- surely the reasons relate to one or more of the above, and there should be some justification to that end.

One guess of mine is also that two adjacent layers, one E-W and one N-S would be fairly minimal in cross-talk due to the perpendicular nature of the traces, versus two adjacent layers where the layers are routed free-style. Would you agree?

\$\endgroup\$
  • \$\begingroup\$ Related: electronics.stackexchange.com/q/79145/2028 \$\endgroup\$ – JYelton Jun 28 '19 at 20:34
  • 1
    \$\begingroup\$ In your second paragraph, compared to what? Neater than what? More layers than what? Your statements about crosstalk assume that some signals are intentionally routed in perpendicular directions but that is not inherent in Manhattan routing. You are leaving too much unspoken to get a good answer. \$\endgroup\$ – Elliot Alderson Jun 28 '19 at 20:35
  • \$\begingroup\$ @ElliotAlderson, updated to hopefully provide more clarity into what I'm trying to gain insight on. \$\endgroup\$ – Kirill Safin Jun 28 '19 at 20:38
  • \$\begingroup\$ Freestyle routing sounds to me as if the designer doesn't know what he/she is doing. But using Manhattan-style routing or just using a ground plane "because everyone does it",will not necessarily improve the design if you still don't know what you're doing. \$\endgroup\$ – Huisman Jun 28 '19 at 21:03
  • 2
    \$\begingroup\$ I often start routing manhattan-style, however, by the time I finish you wouldn't be able to tell! \$\endgroup\$ – evildemonic Jun 28 '19 at 21:34
10
\$\begingroup\$

The primary advantage of a Manhattan route is that it can always be completed. You just need to have enough board area to accommodate all of the traces — but otherwise, you'll never find yourself unable to complete a route. This can be important if you need to get a layout done on a fixed schedule — the amount of work is roughly proportional to the number of pins, and you won't spend days or weeks trying to complete the task because of blockages.

Other routing algorithms might be more efficient than Manhattan in terms of board area and the number of vias required (saving money on each board produced), but they cannot guarantee completion of the route in every case, which means that there's a nonrecurring cost risk in terms of the engineering effort required.

|improve this answer|||||
\$\endgroup\$
  • \$\begingroup\$ That makes a lot of sense. Then if we take density and efficiency aside, are there any reasonable drawbacks to Manhattan? \$\endgroup\$ – Kirill Safin Jun 28 '19 at 21:33
  • 1
    \$\begingroup\$ @KirillSafin: Yes, the routes will always be their Manhattan length at a minimum, and possibly longer. Other algorithms will be able to make some (many?) routes shorter than that. \$\endgroup\$ – Dave Tweed Jun 28 '19 at 22:02
  • \$\begingroup\$ In 1976 we did at least 1 PCB design a week with no simulators or tests with 30 or more DIPS in Manhattan style. Vias were free compared to the cost of copper and volume cost reductions. CMOS was slow rise time so adjacent trace crosstalk was rarely an issue. fast forward to the present with rise times reduced by x100 and dynamic currents now from 25 Ohm switches with 74ALC’ devices vs 300~1200 OHm CD4000 series CMOS switches thus producing 1000x more crosstalk and traces now routinely 3 mil track and gap instead of 20 mil increases crosstalk another order of magnitude. Manhattan Noise!!! \$\endgroup\$ – Tony Stewart Sunnyskyguy EE75 Jun 29 '19 at 0:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.