Trace Length Matching of High Speed Interfaces on STM32MP257

JHardestEE · ‎2025-03-12

Dear ST Rep,

The information within AN5489 regarding the pin-lengths for the high-speed serial interfaces of the MP2 (ie,. DSI / CSI / LVDS / and "COMBOPHY") potentially has some errors, or at the very least is unclear and I would greatly appreciate some help making sense of it.

Side note - The raw pin-length data provided for the DDR interface is exactly what a designer wants/needs in a design. I'm not sure why the same information for the other high-speed interfaces on the MP2 was converted as it was, but it's far less useful (imho). If the raw pin length data could be provided for the above-mentioned interfaces - all my questions will be answered.

To that end - I struggle to make sense of the information regarding the trace length matching for high-speed interfaces (via AN5489), for the reasons stated below:

Per Section 8.19: (pg. 78; Rev3)
"Each package has been optimized to provide easier length matching when differential balls pair signals are not
directly on adjacent balls. Example: package with 0.8 mm ball pitch, when differential pairs are on two different
rows, the package already have around 800 μm length internal difference to allow the PCB track to match total
length with minimum or even no additional routing complexity. Following table shows (for example, xN minus xP)
length difference (inside package) at ball level to be taken into account by the PCB tool."
Per Figure 47 one sees that: (pg 78; Rev3)

The Grey Diff. Traces come from pads at the same row, therefore the PCB routed traces will already be equal and no additional tuning is required.

Conclusion - GOOD, this agrees with the statements in Section 8.19.

The Green Diff. Traces have the "xN" pad on the inside row and the "xP" pad on the outside row. Thus, as the arrow indicates (and Section 8.19 states) "xN minus xP" = -800um (ie, pin length on the outside pad is longer than that of the inside pad, mitigating the length tuning needed on the PCB).

Conclusion - GOOD, this agrees with the statements in Section 8.19.

The Orange Diff. Traces have the "xP" pad on the inside row and the "xN" pad on the outside row. Thus, as the arrow indicates (and Section 8.19 states) "xN minus xP" = 1131um (ie, pin length on the outside pad is longer than that of the inside pad, mitigating the length tuning needed on the PCB).

Conclusion - GOOD, this agrees with the statements in Section 8.19.

The Blue Diff. Traces have the "xP" pad on the inside row and the "xN" pad on the outside row. Thus, based on the arrow (and Section 8.19 statements) "xN minus xP" should equal 800um (ie, pin length on xN is greater than that on xP, to mitigate length tuning on the PCB), but Fig. 47 shows this difference as being (-)800um.

Conclusion - This is the exact opposite of the statements in Section 8.19, this exacerbates the length tuning required on the PCB.

I'm guessing that the (-) sign is an error, and that the value should be (+) 800um (ie,. xN has a longer pin length than xP). Yes - ?
OR - Should the arrow be pointing the other direction (xP minus xN); and really it's the shorter pin length that's always subtracted from the longer pin-length? If this is true, then the orange diff traces appear to be in the reverse order too, but somehow still got the correct pin-length of xN minus xP.

Reference image of Figure 47:

If the (-) sign on the Blue Diff. Traces is simply a typo, and the logic is simply "xN minus xP = difference length", then everything above makes sense. However, if we do so, and then apply said logic to the very next item in the app note "Table 25. Package length matching values", we find the following errors/discrepancies with the statements in Section 8.19 (Each package has been optimized to provide easier length matching when differential balls pair signals are not directly on adjacent balls).

Pin Name	Pad/Ball	Length Delta	Conclusion / Notes:
DSI Interface:
DSI_CKP	B4	207	Therefor: Pin length of A4 (the Outer pin) is 207um longer than B4s (the Inner pin)
DSI_CKN	A4	Conclusion: This agrees with the statements in Section 8.19.
DSI_D0P	B5	57	Therefor: Pin length of A5 (the Outer pin) is 57um longer than B5s (the Inner pin)
DSI_D0N	A5	Conclusion: This agrees with the statements in Section 8.19.
DSI_D1P	B6	-366	Therefor: Pin length of A6 (the Outer pin) is 366um Shorter than B5s (the Inner pin)
DSI_D1N	A6	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
* There are no information, regarding the length tuning for DSI Diff Pairs: DSI_2P/N & DSI_3P/N. Is this missing?
Pin Name	Pad/Ball	Length Delta	Conclusion / Notes:
CSI Interface:
CSI_CKP	C1	147	Therefor: Pin length of C1 (the Outer pin) is 147um Shorter than C2s (the Inner pin).
CSI_CKN	C2	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
CSI_D0P	E2	-450	Therefor: Pin length of E1 (the Outer pin) is 450um Shorter than E2s (the Inner pin).
CSI_D0N	E1	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
CSI_D1P	D1	162	Therefor: Pin length of D1 (the Outer pin) is 162um Shorter than D2s (the Inner pin).
CSI_D1N	D2	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
Pin Name	Pad/Ball	Length Delta	Conclusion / Notes:
USB Interface:
USBH_HS_DP	W11	-11	Therefor: Pin length of W11 (the Outer pin) is 11um longer than V11s (the Inner pin).
USBH_HS_DM	V11	Conclusion: This agrees with the statements in Section 8.19.
USB3DR_DP	W12	27	Therefor: Pin length of W12 (the Outer pin) is 27um Shorter than V12s (the Inner pin).
USB3DR_DM	V12	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
Pin Name	Pad/Ball	Length Delta	Conclusion / Notes:
COMBOPHY Interface:
COMBOPHY_TX1P	U15	336	Therefor: Pin length of V15 (the Outer pin) is 336um longer than U15s (the Inner pin).
COMBOPHY_TX1N	V15	Conclusion: This agrees with the statements in Section 8.19.
COMBOPHY_RX1P	U16	-191	Therefor: Pin length of V16 (the Outer pin) is 191um Shorter than U16s (the Inner pin).
COMBOPHY_RX1N	V16	Conclusion: This is worse for PCB length tuning and Contrasts with the statements in Section 8.19.
Note: LVDS interface not included for brevity…

Given how many conflicts there are within Table 25, I have to assume that the length deltas are not simply "xN minus xP". That said, if the length delta was instead a measure in favor of the expected PCB trace length matching, I don't understand why there are both positive and negative values.(?)

Seeing that the routing guidelines for many of these signals is to maintain +/- 5mils, wrongly interpreting the above information will quickly put the routing out of tolerance and potentially nonfunctional.

Apologies for my ignorance here but any assistance or clarification as to how to correctly implement trace length matching for these interfaces would be greatly appreciated. :)

Thank you for you time and efforts and have a great day.

~ Jay

PatrickF · ‎2025-03-13

HI @JHardestEE

You question make sense and we already got similar from other customers.

I agree the description is somewhat fuzzy or confusing.

The AN5489 is already planned to be updated in an upcoming release.

The paragraph:
"Each package has been optimized to provide easier length matching when differential balls pair signals are not directly on adjacent balls. Example: package with 0.8 mm ball pitch, when differential pairs are on two different
rows, the package already have around 800 μm length internal difference to allow the PCB track to match total length with minimum or even no additional routing complexity. Following table shows (for example, xN minus xP) length difference (inside package) at ball level to be taken into account by the PCB tool."

will be simplify into:
"When possible, each package has been optimized to provide easier length matching when differential balls pair signals are not directly on adjacent balls. Following table shows internal package track length difference xP minus xN/xM at ball level to be taken into account by the PCB tool."

Note the correction of "xN minus xP" in "xP minus xN/xM"

Drawing will be corrected accordingly (sign will changes for the green and the yellow ones).

example, a negative value mean the internal track length of P is longer than internal track length of N (e.g. ideal value for the blue).This mean the PCB track length of P will be shorter than the PCB track length of N.

On some packages, the package inner length was not ideal due to other constrains (so might need more length compensation on PCB), this is why you could see some values not according to drawing (in length and in sign).

Anyway, for differential routing, we have seen that is is more important to route a pair in a good differential way than focus on very strong alignment of length (+/-5 mils is quite challenging).

Furthermore, DSI data were missing for D2 and D3, here are the values:

As a working reference, you could also look at STM32MP257F-DK or STM32MP257F-EV1 design files.

Regards,

In order to give better visibility on the answered topics, please click on 'Accept as Solution' on the reply which solved your issue or answered your question.