LMK04828: T_skew

Hiroshi-san,

The primary factors that affect the skew between outputs on different devices are:

• On-die routing length differences from clock distribution mux to channel path - should be minimal for the same output on two different devices
• Process variation, fast or slow corners, affecting switching speed of retiming flip-flops in the channel divider - big source of variation between devices
• Temperature variation between devices, affecting switching speed of retiming flip-flops in the channel divider and drive strength of output format buffers - devices kept at similar temperature will have less skew than devices at dissimilar temperatures
• Voltage variation between devices, affecting switching speed of retiming flip-flops in the channel divider and (to a limited extent) drive strength of output format buffers - devices kept at similar voltage will have less skew than devices at dissimilar voltages (effect is less than temperature because of on-die LDOs and current regulators designed to keep most circuits behaving similarly independent of small supply voltage variations)
• Different output formats, affecting rise/fall times and delay through the specific format's driver transistors - devices with the same output format on the same CLKout should have less skew variation due to other effects such as temperature and voltage, compared with devices with different output formats on the same CLKout
• Aging, which looks like process variation over long-term - devices with similar lifetime powered hours, operating at similar temperatures, will tend to age similarly. Aging effects take a very long time (years or decades) to observe even minor effects at maximum operating temperature, so this generally isn't a concern, but it's on the list for completeness.

It turns out that jitter has no overall effect as long as it's random, since skew measurements can be made against average values instead of instantaneous values to mostly ignore any cycle-to-cycle variation. Jitter from spurs may not be random, and so this analysis can get more complicated. But most of the time the clock outputs should not have significant enough spurious jitter contribution to affect skew on the picosecond scale.

Some of these parameters will be fixed during device usage (routing length, process variation), while some parameters can vary between power cycles on different devices (output format, temperature, voltage, aging). If variable parameters are kept consistent between power cycles, measured skew differences should be the same (deterministic).

To answer your questions:

1. As long as the variable parameters are kept consistent between the two devices, you can reasonably expect to measure the same average 20ps difference every time.
2. The measured value will change if you change any of the parameters described above, so different format, different device (process), different CLKout, etc will all yield different measured values. However, if you repeat this new measurement across power cycles while keeping all new parameters fixed, the value should be consistent as before.

Regards,

Derek Payne

Hi Derek san,

Thank you for the very detailed commentary and answers.
I understand well.
I will report to the customer.

Regards,

Hiroshi