LMK00301: Clock synchronization scheme

Part Number: LMK00301
Other Parts Discussed in Thread: LMK00105

Dear Experts,

Recently, a client has a project that requires the use of LMK00301SQE/NOPB and LMK00105SQ/NOPB for clock synchronization among multiple devices. The specific connection diagram is as follows:

eca0b15d-232c-445e-b5cc-af77859b4a76.gif
DSP requirements for input clock specifications:
(1) Level range: 1.26V-1.8V
(2) Required duty cycle range: 48%-52%
(3) The frequency deviation is preferably controlled within 10 ppm
(4) The clock frequency deviation of DSP among multiple service boards should be minimized

The customer currently has the following application-related questions that require consultation, and hopes to receive answers from experts. 

1. After cascading LMK00301SQE, how are the device tPD and P2P skew superimposed? According to this scheme, what is the total skew upper limit of the entire clock chain? Will the channel skew and duty cycle distortion of LMK00105SQ input and output itself affect the duty cycle specifications of DSP?
2. Is there any recommended differential impedance, differential pair length tolerance, and equal length constraint for clock routing between different service boards for cross-backplane LVPECL transmission?
3. Is there a recommended configuration for LMK00105SQ power supply filtering, specifically a coupling scheme for LVPECL input and LVCMOS output? Does the 1.8V LVCMOS output terminal require series resistance matching to ensure that the output level falls within the DSP's required range.
4. The input and output of LMK00105SQ belong to cross-power domain conversion. Does the chip itself have power-on timing requirements? If the buffer is powered on earlier than the DSP, is there a risk of reverse leakage in the 1.8V clock output pin of LMK00105SQ when the DSP power supply is not powered on?

Thank you

  • Hey Colin,

    I'm taking a look at these specs, I should have an update for you tomorrow.

    Thank you,

    Tristan

  • Hey Colin,

    I just want to clarify as I look at the prop delay here-- is the below an accurate depiction of the connections?

    If so, I can start stacking up the skew and prop delay to get you an approximate skew.

    1.) We do not spec a part to part skew for LMK00301, but in general you can follow the below rules in 1a and b. Skew should not affect duty cycle, but once you confirm the connections of the devices I can look at how the DCD will affect, as well as a total skew number.

    1a) Prop delay is summed, including trace or cable delays-- usually cable delays will dwarf the rest of the system

    1b) Skew is a measure of the maximum difference in delay, easiest to calculate as sum of devices and traces in one path given maximum tPD, then calculate another path with minimum and check the difference

    2.) You should follow standard differential impedance, length matching, and differential pairing for LVPECL. For LVPECL, these should be 100 ohm differential (50 ohm single ended) length matched traces. Additionally, to reduce skew, you should match all similar traces (IE all 301->105 connections) to each other.

    3.) Yes, the recommended power supply filtering is below for LMK00105, also available in section 9 of our datasheet, p. 20. As far as the coupling between LMK00301 LVPECL output and LMK00105 input, I would recommend following Figure 9-12 of the LMK00301 datasheet, as seen below.

    3a.)

    3b.) 

    4.) I don't see any powerup timing requirements or leakage current on the datasheet, but I will check in with our team.

    Thank you,

    Tristan

  • Dear Tristan,

    Thank you for your reply. Regarding this application, I have the following questions. Could you please help me answer them? Thank you

    1. According to this scheme, the clock signal needs to be transmitted over long distances across boards and cabinets. Will long-link transmission degrade the clock edge, ultimately causing frequency offset issues in the terminal DSP?
    2. The input duty cycle range of the terminal DSP chip is required to be 48%-52%. What is the output duty cycle specification range of the TI-LMK00105SQ chip, and is there any risk involved?
    3. Can the LVCOMS output clock from the REFout pin of the TI-LMK00301 chip be used as a clock of the same origin as the LVPECL output? Is the output clock from the REFout pin stable?
    4. The power supply for the LVCOMS output of the TI-LMK00301 chip is 3.3/2.5V, while the clock pin level requirement for the DSP is 1.26V~1.8V. Is there a risk in using the power supply after voltage division through a resistor for the DSP?

  • Hey Colin,

    1.) There will be some degradation, as this is a voltage driver, but I don't know if it will be enough to cause frequency offset issues in your DSPs. How long is the cable between boards?

    2.) The duty cycle of these devices, as buffers, is relative to the input duty cycle. Assuming a 50% input, the LMK00301 will have an output duty cycle between 45 and 55%, before the LMK00105 buffers. The LMK00105, although not specified, should be a maximum duty cycle distortion of +-5%, meaning that your worst case duty cycle for one of these cases is +-5% for each buffer in the path. This means, if my block diagram above is correct, your worst case is 30% or 70% duty cycle(301M1>301M2>301S>105).

    3.) Yes, this should be a same source clock as the lvpecl outputs. The output should be stable, as specified in the LMK00301 datasheet, in the section of electrical characteristics labeled LVCMOS Output. To note, the propagation delay of the LVCMOS REFout is significantly different than the LVPECL prop delay: Typ ~1.5ns vs ~360ps.

    3a.)

    4.) I believe you mean using the LMK00105 to drive the DSP? In that case, the driver should be able to handle a resistor divider from 3.3V to 1.65V, specifically using a 50 ohm to gnd termination resistor.

  • Hi Tristan,

    1. According to the previous reply, referring to the LVPECL AC coupling design in Figure 9-12 of the LMK00301 data sheet, we are dealing with long-distance cross-board transmission. Should the 100nf blocking capacitor be placed at the transmitting output end or the receiving end?
    2. The output of the LMK00105SQ chip is LVCMOS, which is connected to the DSP clock input. Is it sufficient to simply connect a 0Ω resistor in series?
    3. The current scheme involves multiple stages of cascading, with a deviation of ±5% at each stage. Is this deviation due to chip batch differences or influenced by the link? If it is influenced by the link, can the duty cycle be optimized by adjusting the LVCMOS series resistance or other methods? How can the consistency between different slots be ensured?
    4. Are there any devices with smaller duty cycle deviations?

    Thank you

  • Hey Colin,

    1.) The 100nf capacitor should be before any rebiasing or termination of the signal. Past that, it doesn't really matter if it's close to the receiver or transmitter, as long as your terminations and re-biasing are close to the receiver.

    2.) Given the current block diagram, specifically running the 00105 outputs at 1.8V, it should be perfect to have the series zero ohm, just make sure that the output transmission lines are appropriately 50 ohm impedance matched.

    3.) The deviation is due to process variation, but is also a spec across temperature. The typical duty cycle distortion will be less, but we spec this over the entire temperature range and across process of the device. Supply can play some role here, but not as much as temperature and process.

    4.) We do have devices with lower duty cycle distortion, however once you cascade these devices you have the possibility of stepping past the 48-52% you need. I'm also not sure how well they will work in your application, as they are LP-HCSL devices (0-750mV swing), specifically the LMKDB family of parts.

     

    As we look at the duty cycle distortion, what is the possible output duty cycle of your crystal oscillator? This will be what your distortions start from, so if you are already at 45-55% from the crystal, it may not be possible to reach the 48-52% regardless.

    Thank you,

    Tristan