Circular arcs. Scott McMorrow Teraspeed Consulting Group LLC 121 North River Drive Narragansett, RI 02882 (401) 284-1827 Business (401) 284-1840 Fax http://www.teraspeed.com Teraspeed® is the registered service mark of Teraspeed Consulting Group LLC On 7/20/2011 10:54 AM, Jack Olson wrote: > I just realized that was a really dumb question. > Can I retract that? > How can you route without bends? > I'm an idiot... > sorry. > > On Wed, Jul 20, 2011 at 9:52 AM, Jack Olson <pcbjack@xxxxxxxxx > <mailto:pcbjack@xxxxxxxxx>> wrote: > > Thanks very much, Scott > > You have tools and knowledge that I never encounter here at my job > (our designs have different constraints than many "SI-Listers", so > I shouldn't have spewed my "real world" comment. Sorry) > but... > > I would like to save this discussion for future reference, and > since we are closer to nailing this down, could you make one > further comment about 45 degree bends if anything needs to be said > about that? I don't think I've ever truly used a 90 degree bend in > trace routing yet (24 years) although I can see 90 degree features > in almost every design (like vias and exiting from rectangular > pads, for example), but even our serpentine tuning uses round > apertures and rounded bends (Mentor). > > That's why I had the assumption that talking about square > apertures and 90 bends (in trace routing) was only a mathematical > exercise > > Since it is difficult for our CAD system to make "the optimal RF > miter", we would probably use two 45 degree bends. Does anything > need to be said about using round apertures with 45 degree bends? > That's what I've seen in most board trace routing, except for the > older tape-up designs. > > Thanks again for sharing your work, > Jack > > > . > > On Wed, Jul 20, 2011 at 7:58 AM, Scott McMorrow > <scott@xxxxxxxxxxxxx <mailto:scott@xxxxxxxxxxxxx>> wrote: > > Jack > > I've decided to put some numbers to the Right Angle Bend > discussion, since it's nice to know when we need to be > concerned about a particular physical phenomena. > > Your point regarding round apertures on traces is important, > although slightly misleading. A round aperture is applied to > the end of a trace with the center of the radius at the center > of the trace width, such that the outside apex is rounded, > while the inside apex is square. Thus, a trace corner with > circular end apertures still has excess capacitance when > compared to the optimal RF miter[1]. If we use Bogatin's > approximation of a 50% miter (corner sliced diagonally in the > center at 45 degrees, which is not quite correct, but is > close) to compute the excess capacitance, then using a > circular end aperture on traces at corners reduces capacitance > by (1-pi/4), or about 21%, which helps, but still leaves us > with excess capacitance. > > Here are some calculations based on Bogatin, "Signal Integrity > Simplified", page 317, which I've verified: > > Square corner capacitance(pf) = 40/Z0 x sqrt(Er) x w > where w = width in inches > Er = dielectric contstant > Z0 = characteristic impedance of trace > > 50 ohm impedance, Er = 4 > > 10 Gbps (5 GHz Nyquist) with square apertures. > Case 1) A 100 mil wide trace - 160 fF = 199 ohm shunt Z at 10 > GHz, produces a 40 ohm tdr blip for 17 ps (17% UI). > Case 2) A 10 mil wide trace - 16 fF = 1990 ohm shunt Z at 10 > GHz, produces a 48.8 ohm tdr blip for 1.7 ps (1.7% UI). > Case 3) A 5 mil wide trace - 8 fF = 3980 ohm shunt Z at 10 > GHz, produces a 49.4 ohm tdr blip for 850 fs (0.85% UI). > Case 4) A 1 mil wide trace - 1.6 fF = 19900 ohm shunt Z at 10 > GHz, produces a 49.75 ohm tdr blip for 170 fs (0.2% UI). > > 20 Gbps (10 GHz Nyquist) with square apertures. > Case 1) A 100 mil wide trace - 160 fF = 99 ohm shunt Z at 10 > GHz, produces a 33 ohm tdr blip for 17 ps (34% UI). > Case 2) A 10 mil wide trace - 16 fF = 995 ohm shunt Z at 10 > GHz, produces a 47.6 ohm tdr blip for 1.7 ps. (3.6% UI). > Case 3) A 5 mil wide trace - 8 fF = 1990 ohm shunt Z at 10 > GHz, produces a 48.8 ohm tdr blip for 850 fs (1.7% UI). > Case 4) A 1 mil wide trace - 1.6 fF = 9947 ohm shunt Z at 10 > GHz, produces a 49.75 ohm tdr blip for 170 fs (0.3% UI). > > 20 Gbps (10 GHz Nyquist) With circular end apertures on traces > (reduce excess shunt capacitance by 21%) > Case 1) A 100 mil wide trace - 126 fF = 126 ohm shunt Z at 10 > GHz, produces a 36 ohm tdr blip for 17 ps (34% UI). > Case 2) A 10 mil wide trace - 12.6 fF = 1265 ohm shunt Z at 10 > GHz, produces a 48 ohm tdr blip for 1.7 ps (3.4% UI). > Case 3) A 5 mil wide trace - 6.3 fF = 2530 ohm shunt Z at 10 > GHz, produces a 49 ohm tdr blip for 850 fs (1.7% UI). > Case 4) A 1 mil wide trace - 1.26 fF = 12650 ohm shunt Z at 10 > GHz, produces a 49.8 ohm tdr blip for 170 fs (0.3% UI). > > > So, for real traces, on real boards, with real CAD end > apertures, a single corner on traces that are 1 mil to 10 mil > wide, at frequencies of 10 GHz or below, neffectively cannot > be measured without specialized de-embedding methods. The > discontinuity is too small and too short. For microwave width > traces, those corners are large discontinuities with real > consequences. > > If we happen to be running at 56 Gbps (28 GHz Nyquist) > Case 2) A 10 mil wide trace - 12.6 fF = 452 ohm shunt Z at 10 > GHz, produces a 45 ohm tdr blip for 1.7 ps (9.5% UI). > Case 3) A 5 mil wide trace - 6.3 fF = 604 ohm shunt Z at 10 > GHz, produces a 46 ohm tdr blip for 850 fs (4.8% UI). > Case 4) A 1 mil wide trace - 1.26 fF = 4518 ohm shunt Z at 10 > GHz, produces a 49.5 ohm tdr blip for 170 fs (0.9% UI). > > Conclusions: > > * There is nothing wrong with the information about square > corners found in the RF and Microwave literature. It > applies to the wide, low loss traces used in RF design, > and is quite correct, corners do matter. > * For digital designs traces of 10 mil width or less are > generally used, and as such corner design is essentially > unimportant, due to the benefit of size scaling on > reduction in the size and duration of the discontinuity. > * As a figure of merit, I'd suggest we use 5% of UI, and > +/- 5% Z0 as the point that we start considering that a > particular type of discontinuity matters for SerDes > channels. This is conservative, but I like being > conservative, and working with positive margin, rather > than fighting against negative margin. > * Up to 20-ish Gbps, 90 degree CAD layout corners on > traces up to 10 mil in width have no significant impact > on the signal. > * At 56 Gbps, 90 degree CAD layout corners become > significant starting at 5 mil width. > * At 28 Gbps and 56 Gbps designers may be fighting high > copper and dielectric losses, forcing the usage of low > loss dielectrics (tanD < 0.005), smooth copper foils > (surface roughness < 0.4 micron RMS), and wider traces. > In these cases, corner mitering, or routing of traces > using circular arcs, might be considered, when traces > are wider than 10 mil @ 28 Gbps and 5 mil @ 56 Gbps. > > > Another way to look at corners is from the view of a > transmission line with distributed loading. On page 317 of > Eric Bogatin's book, he shows his geometric derivation for the > computation of corner capacitance. We can reduce his geometry > into 3 square cells at a corner, which can be further divided > into 6 equal sized triangular sections. One triangular > section represents the excess corner capacitance. If we call > this the "unit corner cell" then it's clear that corner > capacitance represents an increase of the capacitance of the > entire cell by 6/5, or 1.2. Knowing this we can compute the > distributed Impedance of the line as Z(distributed) = Z0 x > sqrt(5/6) = 0.91 Z0. For 50 ohm impedance, that amounts to a > distributed transmission line impedance of 45.6 ohms. So, if > we were to cascade corners in a diagonal stairstep fashion, > the distributed impedance of the line would asymptotically > approach around 45 ohms, since the capacitance becomes > averaged across a longer distance. This is independent of the > line width. In addition, the distributed delay of the > transmission line is increased by sqrt(6/5) or by 1.095, > across the unit cell section, or about 16 ps per inch of > additional delay for Er=4. > > So where could we have a problem? Using the distributed > version of the analysis, we can see that stair step type > structures might accentuate corner capacitance issues. One > place where this can happen is in the routing escape pin > fields of BGA and connectors. Another is the increase in > actual delay that can be seen in trace delay matching > serpentine sections. Even if coupling between serpentine > sections is minimized, a unit rectangular serpentine section > has 4 corners. For 5 mil trace width, each corner has an > excess capacitance of 6.3 fF, an incremental delay adder of > about 300 fs. For the 4 corners of a serpentine section, > there is 1.2 ps of additional delay, not accounted for by net > length calculations. Short squarish delay matching sections > (so called delay bumps or blips along a line) with 4 corners > per blip will experience a large percentage increase in delay > over the net length calculation vs. longer rectangular delay > matching sections, with fewer total number of corners. > > Real world? It depends on the world you're designing in. > > best regards, > > Scott > > > 1. Experimental derivation of optimal miter can be found in: > R. J. P. Douville and D. S. James, Experimental study of > symmetric microstrip bends and their compensation; IEEE Trans. > Microwave Theory Tech., vol. MTT-26, pp. 175-182, Mar. 1978. > > > > Scott McMorrow > Teraspeed Consulting Group LLC > 121 North River Drive > Narragansett, RI 02882 > (401) 284-1827 <tel:%28401%29%20284-1827> Business > (401) 284-1840 <tel:%28401%29%20284-1840> Fax > > http://www.teraspeed.com > > Teraspeed® is the registered service mark of > Teraspeed Consulting Group LLC > > > On 7/18/2011 10:14 AM, Jack Olson wrote: >> Speaking from the board designers point of view: >> I may never understand why people use square corners in their models >> when >> they want to experiment with this subject. Are we talking about 90 >> degree >> bends? or drawing traces with square corners? Maybe it is >> interesting from a >> mathematical point of view, but I've never met a board designer who >> used >> square apertures to draw traces. Every modern board design I've ever >> seen >> (by modern I mean designed in a CAD system , not the old manual tape >> layouts) uses round apertures, and when using a round aperture the >> width is >> always constant, no matter how you "bend" the trace. >> Furthermore, I haven't seen too many designs where the bends weren't >> 45 >> degrees anyway, why would anyone route 90 degree bends? longer runs, >> in most >> cases.... but theoretically I suppose its interesting to discuss. >> Real world? not so much. >> >> onward thru the fog, >> Jack >> >> >> . >> Date: Fri, 15 Jul 2011 16:56:38 -0400 >> From: Scott McMorrow<scott@xxxxxxxxxxxxx> >> <mailto:scott@xxxxxxxxxxxxx> >> Subject: [SI-LIST] Re: Right Angle Bends >> >> Lee >> >> It's not so much "misinformation" but "misapplication" of >> information. >> The "no right angle bend rule" makes perfect sense on RF microstrip >> boards where the traces are about 1 or 2 mm from the plane, and are >> extremely wide. When you're dealing with 50 ohm traces that are 100 >> mils wide, excess capacitance at the corner is a big deal, since the >> discontinuity acts for about 140 mils in distance (the diagonal >> across >> the corner). If built on an FR4-type material with a Dk = 4, this >> corner discontinuity has about a 25 ps time duration, which can be >> significant. Additionally, RF engineers are always trying to reduce >> narrow band return loss to extremely low levels, far beyond that >> which >> is necessary for broadband digital. >> >> However, when we scale the traces down to 5 mil width, the >> discontinuity >> is same relative size, but 1/20th the physical size, and acts for >> only >> about 7 mils, or about 1.25 ps. IF a 1.25 ps discontinuity is a big >> deal to an engineer then this might matter, or if you have 20 >> corners, >> the cumulative effect of this discontinuity will span a duration of >> 25 >> ps, or 1/4 of a 10 Gbps bit time. But 1.25 ps of excess capacitance >> does not matter until we designing for 50 Gbps, and only a fool would >> route a trace with 20 corners in it, so effectively corners are not >> an >> issue. >> >> Lee, you are correct. >> >> Scott >> >> Scott McMorrow >> Teraspeed Consulting Group LLC >> 121 North River Drive >> Narragansett, RI 02882 >> (401) 284-1827 <tel:%28401%29%20284-1827> Business >> (401) 284-1840 <tel:%28401%29%20284-1840> Fax >> >> http://www.teraspeed.com >> >> Teraspeed® is the registered service mark of >> Teraspeed Consulting Group LLC >> >> >> On 7/15/2011 3:15 PM, Lee Ritchey wrote: >>> That is another urban legend. Never been true. I've seen >>> fabricators say >>> this and then etch outer layers with all sorts of surface mount >>> pads that >>> have traces entering them that result in right angle corners and >>> never >>> complain! >>> >>> We'll probably all go to our graves before we flush out all this >>> misinformation! > > > ------------------------------------------------------------------ To unsubscribe from si-list: si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field or to administer your membership from a web page, go to: //www.freelists.org/webpage/si-list For help: si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field List technical documents are available at: http://www.si-list.net List archives are viewable at: //www.freelists.org/archives/si-list Old (prior to June 6, 2001) list archives are viewable at: http://www.qsl.net/wb6tpu