Physical layer devices (PHYs) may include multiple Serializer/Deserializer (SerDes) lanes. Each SerDes lane may include a serializer block and a deserializer block. The serializer block generally converts data from a parallel format to a serial format. The deserializer block generally converts data from a serial format to a parallel format.
According to an example embodiment, a circuit on a chip for serial data applications may comprise a common phase-locked loop (PLL). The common PLL may be configured to receive an off-chip reference clock signal generated external to the chip and produce an on-chip reference clock signal. The on-chip reference clock signal may be higher in frequency relative to the off-chip reference clock signal. The circuit may comprise a plurality of serializer/deserializer (SerDes) lanes each composed of a respective transmitter, receiver, and fractional-N (frac-N) PLL. The on-chip reference clock signal may be distributed to each respective frac-N PLL for use in generating a respective output clock signal with a respective frequency. The respective transmitter and receiver of each SerDes lane may be configured to operate at respective transmit and receive data rates. The respective transmit and receive data rates may be based on the respective frequency and may be independent of data rates of other SerDes lanes of the plurality of SerDes lanes.
The off-chip reference clock signal may be configured to operate in a low frequency range of 100 MHz to 156 MHz and the on-chip reference clock signal may be configured to operate in a higher frequency range of 400 MHz to 600 MHz.
The circuit may be configured to suppress a respective modulation noise, generated by each respective frac-N PLL, by increasing frequency of the respective modulation noise to enable each respective frac-N PLL to filter the respective modulation noise.
Each respective frac-N PLL of the plurality of SerDes lanes may be composed of a respective divider and sigma-delta modulator. Each respective sigma-delta modulator may be configured to modulate, randomly, a respective divide value of the respective divider.
The common PLL may be an integer PLL with a multiplying factor and the multiplying factor may be configured to suppress an initial portion of quantization noise introduced by modulating each respective divide value of each respective divider of each respective frac-N PLL of the plurality of SerDes lanes.
Each respective frac-N PLL of the plurality of SerDes lanes may be further composed of a respective out-of-band parasitic pole configured to suppress a remaining portion of the quantization noise. The remaining portion may be at higher offset frequencies relative to offset frequencies of the initial portion. The initial portion and the remaining portion may be suppressed below a noise level of noise generated from intrinsic phase detector and voltage controlled oscillator (VCO) noise sources on the chip.
The integer PLL may be a first integer PLL and each respective out-of-band parasitic pole may be configured high enough to not affect closed-loop behavior of the respective frac-N PLL. Each respective out-of-band parasitic pole in combination with configuration of the multiplying factor may enable jitter and phase noise performance of each respective frac-N PLL of the plurality of SerDes lanes to be comparable to that of a second integer PLL that employs a same integer divide value as the respective frac-N PLL.
Each respective frac-N PLL of the plurality of SerDes lanes may have an order that is lower relative to that of the respective sigma-delta modulator.
The order of each respective frac-N PLL may be two and that of the respective sigma-delta modulator may be three.
Each respective frac-N PLL of the plurality of SerDes lanes may be composed of a respective spread spectrum clock (SSC) pattern generator configured to modulate the respective frac-N PLL, directly, according to a respective SSC profile.
The chip may be a physical layer device (PHY). Each SerDes lane may be configured to employ the respective transmit and receive data rates to communicate with a respective media access controller (MAC). The respective transmit and receive data rates may be configured independent of data rates employed by other SerDes lanes of the PHY for communicating with other MACs.
According to another example embodiment, a method may comprise producing an on-chip reference clock signal from a common phase-locked loop (PLL) on a chip based on an off-chip reference clock signal generated external to the chip. The on-chip reference clock signal may be higher in frequency relative to the off-chip reference clock signal. The method may comprise distributing the on-chip reference clock signal to each respective frac-N PLL of a plurality of serializer/deserializer (SerDes) lanes each composed of a respective transmitter, receiver, and fractional-N (frac-N) PLL. The method may comprise generating a respective output clock signal with a respective frequency from each respective frac-N PLL based on the on-chip reference clock signal distributed. The method may comprise operating each respective transmitter and receiver of each SerDes lane at respective transmit and receive data rates based on the respective frequency and may be independent of data rates of other SerDes lanes of the plurality of SerDes lanes.
The method may comprise operating the off-chip reference clock signal in a low frequency range of 100 MHz to 156 MHz and operating the on-chip reference clock signal in a higher frequency range of 400 MHz to 600 MHz.
The method may comprise suppressing a respective modulation noise, generated by each respective frac-N PLL, by increasing frequency of the respective modulation noise to enable each respective frac-N PLL to filter the respective modulation noise.
Each respective frac-N PLL of the plurality of SerDes lanes may be composed of a respective divider and sigma-delta modulator. The method may further comprise employing each respective sigma-delta modulator to modulate, randomly, a respective divide value of the respective divider.
The common PLL may be an integer PLL with a multiplying factor and the method may further comprise configuring the multiplying factor to suppress an initial portion of quantization noise introduced by modulating each respective divide value of each respective divider of each respective frac-N PLL of the plurality of SerDes lanes.
Each respective frac-N PLL of the plurality of SerDes lanes may be further composed of a respective out-of-band parasitic pole and the method may further comprise configuring each respective out-of-band parasitic pole to suppress a remaining portion of the quantization noise. The remaining portion may be at higher offset frequencies relative to offset frequencies of the initial portion. The initial portion and the remaining portion may be suppressed below a noise level of noise generated from intrinsic phase detector and voltage controlled oscillator (VCO) noise sources on the chip.
The integer PLL may be a first integer PLL and configuring each respective out-of-band parasitic pole may include configuring each respective out-of-band parasitic pole to be high enough to not affect closed-loop behavior of the respective frac-N PLL. Each respective out-of-band parasitic pole in combination with configuration of the multiplying factor may enable jitter and phase noise performance of each respective frac-N PLL of the plurality of SerDes lanes to be comparable to that of a second integer PLL that employs a same integer divide value as the respective frac-N PLL.
The method may further comprise configuring each respective frac-N PLL of the plurality of SerDes lanes to have an order that is lower relative to that of the respective sigma-delta modulator. The order of each respective frac-N PLL may be two and that of the respective sigma-delta modulator may be three.
Each respective frac-N PLL of the plurality of SerDes lanes may be composed of a respective spread spectrum clock (SSC) pattern generator and the method may further comprise configuring each respective SSC pattern generator to modulate the respective frac-N PLL, directly, according to a respective SSC profile.
The chip may be a physical layer device (PHY), and the method may further comprise configuring each SerDes lane to employ the respective transmit and receive data rates to communicate with a respective media access controller (MAC) and configuring the respective transmit and receive data rates independent of data rates employed by other SerDes lanes of the PHY for communicating with other MACs.
It should be understood that example embodiments disclosed herein can be implemented in the form of a method, apparatus, system, or computer readable medium with program codes embodied thereon.
The foregoing will be apparent from the following more particular description of example embodiments, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating embodiments.
A description of example embodiments follows.
In serial data applications it may be useful to have serializer/deserializer (SerDes) lanes (also referred to interchangeably herein as SerDes data lanes) of a physical layer device (PHY) operate over a broad range of data rates and serial protocols. In addition, for maximum flexibility in connecting to external components, it may also be useful to have the PHY connect to multiple logical layer media access controllers (MACs) so that the PHY can be programmed at chip startup to communicate with a particular protocol for a particular application or board design while using a same piece of provider silicon. As such, a PHY with SerDes lanes configured to span a wide range of operational frequencies and protocol requirements may be useful.
Within the PHY, also referred to interchangeably herein as a SerDes PHY, a plurality of SerDes lanes may be configured to communicate with external devices. The SerDes lanes may be composed primarily of a transmitter (TX) and a receiver (RX) to enable such communication. Embodiments disclosed herein enable multiple SerDes data lanes of a PHY to operate across a broad range of diversified data rates from lane to lane without placing difficult constraints with respect to clocking requirements for clocking signal being provided to the SerDes data lanes. An example embodiment relieves a customer who is employing the PHY in their design from adhering to such difficult constraints. For example, the customer may employ low frequency clock signals readily available in their design and are relieved from having to generate a high frequency clock signal that may otherwise be required by the PHY.
According to an example embodiment, each SerDes lane may operate at a completely independent data rate from one another based on a single low frequency reference clock signal input. Achieving such lane-by-lane data rate independence may be enabled by operating each SerDes lane as a fractional-N (frac-N) frequency synthesizer and by using a common phase locked loop (PLL) on the PHY as a frequency step-up PLL for increasing frequency of the single low frequency reference clock signal input.
The use of the step-up PLL avoids having to do any high risk quantization noise cancellation techniques that exist in the art and provides an additional benefit of suppressing a frac-N modulation noise by moving the frac-N noise to higher frequencies where such noise may be filtered by the frac-N PLL. Example embodiments of a fractional modulator, PLL loop dynamics, and additional PLL filtering poles enables a level of the frac-N modulation noise to be suppressed to a point where phase noise and jitter of the frac-N PLL appears identical to that of an integer-N PLL, as disclosed further below.
The off-chip reference clock signal 108 may be configured to operate in a low frequency range of 100 MHz to 156 MHz and the on-chip reference clock signal 110 may be configured to operate in a higher frequency range of 400 MHz to 600 MHz.
The chip 104 may be a physical layer device (PHY). Each SerDes lane 112a-n may be configured to employ the respective transmit and receive data rates to communicate with a respective media access controller (MAC) (not shown) and external devices (not shown). The respective transmit and receive data rates may be configured independent of data rates employed by other SerDes lanes of the PHY for communicating with other MACs.
The PHY 204 is configured to communicate with an outside world 203, such as an external network or external device, via a first interface 205. The first interface 205 may be an interface to a physical medium, such as an optical fiber or copper cable, or may be a wireless interface or any other suitable interface for communicating with external devices.
The apparatus 201 may comprise multiple logical layer media access controllers (MACs), such as the MACs 207a-n, and a second interface 209. The second interface 209 may be a logical multiplexer or media independent interface (MII) disposed between the PHY 204 with the plurality of MACs 207a-n. The plurality of MACs 207a-n may have respective requirements for communication, such as respective requirements for data rates. Each MAC of the plurality of MACs 207a-n may be configured to deliver data between an internal bus 211 of the apparatus 201 and the outside world 203 via a respective SerDes lane (not shown) of the PHY 204. According to an example embodiment, such respective SerDes lanes may be configured to operate at respective transmit and receive data rates that are independent of data rates employed by other SerDes lanes of the PHY 204.
According to an example embodiment of the circuit (not shown) of the PHY 204, such as the example embodiment of the circuit 102 of
Each SerDes lane further includes digital circuitry to interface to a respective logical MAC (not shown) that may be placed by place and route (PNR) tools and referred to interchangeably herein as PNR. For example, the SerDes lanes 312a, 312b, 312c, and 312d include the PNR 322a, 322b, 322c, and 322d, respectively, for interfacing to respective logical MACs (not shown). Included in the COM block 313 is control circuitry that includes the PNR control circuitry 323 that may be used for controlling the PNR 322a, 322b, 322c, and 322d. The COM block 313 further includes a common PLL 306 that may be an integer PLL and COM circuitry 324 that may include miscellaneous circuitry that is common within the circuit 302.
Each SerDes lane includes a TX, RX, and frac-N PLL. For example, the SerDes lane 412a includes the TX 414a, RX 418a, and frac-N 416a, while the SerDes lane 412b includes the TX 414b, RX 418b, and frac-N 416b. Each SerDes lane, that is, the SerDes lane 412a and the SerDes lane 412b, further includes digital circuitry to interface to a respective logical MAC (not shown), that is, the PNR 422a and PNR 422b, respectively. Included in the COM block 413 is control circuitry that may include the PNR control 423 that may be used for controlling the PNR 422a and PNR 422b. The COM block 413 further includes a common PLL 406 that may be an integer PLL as well as the COM circuitry 424 that may include miscellaneous circuitry that is common within the circuit 402.
The LF 538 may be any suitable type of loop filter, such as an integrating loop filter, and the PFD 536 may include a charge-pump (not shown). The output frequency of the output signal 548 may be configured to be Fout=N*Fref, where N is an integer value, such as the integer N 550 that may be combined and input to the divider 534. As such, the PLL 530 may be referred to as an integer PLL. A reset block 552 may be employed to reset the PLL 530.
Alternatively, a pre-scaler R (not shown) may be configured to divide down the reference clock 540 and a post-scaler P (not shown) may be configured to divide down the output frequency of the output signal 548. As such, the output frequency of the output signal 548 may instead be Fout=N*Fref/(R*P) so that some fractional values may be obtained. A stability of an output signal generated by such a PLL, referred to as a fractional PLL, is closely related to a ratio of (Fref/R)/(PLL bandwidth). The PLL bandwidth is often set by other considerations for high performance applications, requiring that R=1. The post-scaler P (not shown) is often set to fixed ratios because the VCO 532 may be operating near a highest rate possible to generate clocks needed to achieve particular data rates. The post-scaler P (not shown) may be used to generate output phases. As such, a frequency resolution of the output signal 548 for data rates in practical applications is generally set by N, which is an integer.
In the graph 600, the VCO noise 650 is the VCO phase noise of the VCO 532 and the detector noise 652 is the phase noise of the PFD 536. LF noise (not shown) of the LF 538 also contributes noise, but has been withheld from the graph 600 so as not to confuse the graph 600. In the example embodiment, it can be assumed that the LF noise (not shown) is set low enough to not be dominant in the PLL 530. A total PLL noise 654 for the PLL 530 corresponds to a sum of the two noise sources, that is, the VCO noise 650 and the PFD noise 652, with respect to the frequency offset 660.
The PFD noise 652 is low pass filtered by the PLL 530, while the VCO noise 650 is high pass filtered. As such, the PFD noise 652 is dominant at low frequencies and the VCO noise 650 is dominant at high frequencies. An integration of the total PLL noise 654 may be converted to total jitter, which, in the example embodiment, is 372 fs root-mean-square (rms). In the example embodiment of the PLL 530, a frequency of the input reference clock signal 540 is assumed to be in a range of 100-156 MHz, which is typical for modern SerDes applications. In order to de-couple a choice of frequency resolution (and therefore data rate resolution) from the reference clock 540, an example embodiment may convert the PLL 530 into a fractional-N synthesizer (also referred to interchangeably herein as a PLL) as disclosed with regard to
In the example embodiment of the PLL 716, a divide value N 750 is modulated between integer values in a way that, on average, a fractional divide value N·F 761 is obtained where N is an integer and F is a fractional value. Modulation of the divide value N 750 may introduce quantization noise. An example embodiment may modulate the divider 734 of the PLL 716 in randomized manner using a sigma-delta modulator 760 to shape the quantization noise so that the quantization noise is high-pass filtered and then low-pass filtered by the PLL 716. The input reference clock signal (REF) 740 may drive the sigma-delta modulator 760 in any suitable manner. For example, the input reference clock signal (REF) 740 may drive the sigma-delta modulator 760, directly, or in a buffered form 762. Further, the input reference clock signal (REF) 740 may be inverted or altered in some other suitable manner to meet timing margins for the PLL 716.
According to an example embodiment, the phase noise response of
In Eq. (1), above, Δ represents a quantization step size (1 divide value step), M is an order of the sigma-delta modulator, where (M=2 in the example embodiment of
In the example embodiment, the quantization noise is reduced by shifting it to higher frequency, where a loop filter, such as the LF 738 of the PLL 716 of
An exact frequency of such a sigma delta reference is not important. Such a frequency may be configured such that it is “high enough” to get sufficient suppression of the quantization noise. Such a frequency may be in the 400-600 MHz range, a frequency that may be distributed easily across a large die for all SerDes lanes from a common multiplier PLL, such as the common PLL 106, the common PLL 306, or the common PLL 406 of
In the example embodiment of
The common PLL 106 may be an integer PLL with a multiplying factor and the multiplying factor may be configured to suppress an initial portion of quantization noise introduced by modulating each respective divide value of each respective divider of each respective frac-N PLL 116a-n of the plurality of SerDes lanes 112a-n. Each respective frac-N PLL 116a-n of the plurality of SerDes lanes 112a-n may be further composed of a respective out-of-band parasitic pole (not shown) that is configured to suppress a remaining portion of the quantization noise. The remaining portion may be at higher offset frequencies relative to offset frequencies of the initial portion. The initial portion and the remaining portion may be suppressed below a noise level of noise generated from intrinsic phase detector (not shown) and voltage controlled oscillator (VCO) noise sources (not shown) on the chip 104.
Each respective out-of-band parasitic pole (not shown) may be configured high enough to not affect closed-loop behavior of the respective frac-N PLL 116a-n. Each respective out-of-band parasitic pole (not shown) in combination with configuration of the multiplying factor may enable jitter and phase noise performance of each respective frac-N PLL of the plurality of SerDes lanes to be comparable to that of an integer PLL that employs a same integer divide value as the respective frac-N PLL, such as disclosed above with regard to
Such modulation is useful because many serial data standards require generation of spread spectrum clocks (SSC). Without frac-N capabilities, it becomes complex to generate SSC profiles and such complex generation may tap multiple VCO output phases or be based on generation of multiple output phases to create such profiles. According to the example embodiment of
Turning back to
The method may check whether a performance of the frac-N PLL is comparable to that of an integer PLL that employs a same integer divide value as the frac-N PLL (1660). If the performance is comparable, the method thereafter ends (1664) in the example embodiment. However, if the performance is not comparable, the method may perform a co-optimization method (1662) with respect to (i) an on-chip frequency reference value (Fref) of a frequency of an on-chip reference clock signal distributed to the frac-N PLL, such as the on-chip reference clock signal 110 of
If yes, the method thereafter ends (1664) in the example embodiment. If no, the method may again employ the optimization method (1662). As such, the method (1653) may iterate an optimization loop 1651 an iterate performing the optimization method (1662) until the performance of the frac-N PLL is comparable to that of an integer PLL that employs a same integer divide value as the frac-N PLL, enabling data rate independence of the SerDes lane relative to other SerDes lanes. By recognizing tradeoffs between choosing an input frequency to a frac-N PLL, an order of a sigma-delta modulator (i.e., ΣΔ), and out of band poles, an influence of quantization noise on the frac-N PLL performance may be minimized or eliminated while employing a frac-N PLL.
It should be understood that the method (1653) may be employed from SerDes lane-to-lane or employed for a single lane and applied to multiple SerDes lanes. For example, selection of input frequency to a frac-N PLL, an order of a sigma-delta modulator (i.e., ΣΔ), and out-of-band poles determined as a result of the optimization loop 1651, may be applied to all SerDes lanes of a circuit, such as the circuit 102 of
Alternatively, local frequency division of the on-chip frequency reference value may be employed with any given SerDes lane and each SerDes lane may employ a respective order of its respective sigma-delta modulator and respective out-of-band poles that are selected for the given SerDes lane independent of selections made for other SerDes lanes. Factors such as power and performance requirements may influence the selection of the input frequency to the frac-N PLL, the order of a sigma-delta modulator (i.e., ΣΔ), and the out-of-band poles that may be employed.
While example embodiments have been particularly shown and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the embodiments encompassed by the appended claims.
The teachings of all patents, published applications and references cited herein are incorporated by reference in their entirety.
While example embodiments have been particularly shown and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the scope of the embodiments encompassed by the appended claims.
This application is a continuation of U.S. application Ser. No. 16/569,445, filed Sep. 12, 2019, which is a continuation of U.S. application Ser. No. 16/352,180, filed Mar. 13, 2019, now U.S. Pat. No. 10,461,917, which is a continuation of U.S. application Ser. No. 15/721,334, filed Sep. 29, 2017, now U.S. Pat. No. 10,291,386. The entire teachings of the above applications are incorporated herein by reference.
Number | Date | Country | |
---|---|---|---|
Parent | 16569445 | Sep 2019 | US |
Child | 17324025 | US | |
Parent | 16352180 | Mar 2019 | US |
Child | 16569445 | US | |
Parent | 15721334 | Sep 2017 | US |
Child | 16352180 | US |