SPI-4.2 dynamic implementation without additional phase locked loops

[0001] This application claims priority to U.S. Application Ser. No. 60/723,458 filed Oct. 4, 2005, which is incorporated herein by reference in its entirety.

[0002] The present disclosure relates generally to network processing devices, and more particularly to a system and method of synchronizing the reception of data on multiple data lanes referenced to a single clock.

[0003] Many integrated circuits today are connected by printed circuit board high-speed parallel data buses. Due to unequal trace lengths on each data lane and other effects, data launched on each lane at the same time tends to arrive at the receiver at slightly different times. As clock rates increase and pulse widths decrease, the ability to deskew the data lanes at the receiver to compensate for the different lane delays becomes more critical. Deskew logic may be static or dynamic. Dynamic deskew logic typically consists of one phase-locked loop (PLL) or delay-locked loop (DLL) for each data lane, which determines the optimal sample time for each data lane. Implementing DLLs for each lane, however, consumes significant integrated circuit area that could otherwise be used for core circuit functions or removed to reduce circuit size and power requirements.

[0010] FIG. 1 illustrates a partial system implementation 100 that operates across a bus system connecting two integrated circuits 102 and 104. The bus system connecting circuits 102 and 104 follows the System Packet Interface Level 4 (SPI-4) Phase 2 Revision 1 implementation agreement OIF-SPI-4-02.1, promulgated by the Optical Internetworking Forum, referred to herein as SPI-4.2. Circuit 102 comprises an SPI-4.2-compliant transmitter 110 and other logic (not shown) for supplying data to the transmitter. Circuit 104 comprises an SPI-4.2-compliant receiver 400 and other logic (not shown) for receiving data from the receiver.

[0011] An SPI-4.2 bus system connects SPI-4.2 transmitter 110 with SPI-4.2 receiver 400. The bus system comprises a 16-bit-wide data bus DAT[15:0], a control line CTL, a data clock DCLK, a two-bit-wide status bus STAT, and a status clock SCLK. DAT, CTL, and DCLK signals originate at SPI-4.2 transmitter 110 and terminate at SPI-4.2 receiver 400. STAT and SCLK originate at SPI-4.2 receiver 400 and terminate at SPI-4.2 transmitter 110. With respect to the transmitter, DATA and CTL are each driven on a differential pair at double data rate, e.g., one symbol on each pair per transition of data clock DCLK. The transmitter is required to maintain a maximum +/−bit time in bit alignment jitter on each data differential pair with respect to DCLK. The transmitter is also required to maintain a maximum +/−bit time relative skew between the data lines.

[0012] Data bus DAT transfers 16 bits of packet data (eight if only one octet remains in a packet) or a control word each clock cycle. When CTL is asserted, DAT contains a control word to be interpreted by SPI-4.2 receiver 400. When CTL is deasserted, DAT contains packet data. Packet data is transmitted in bursts, with a control word immediately preceding and immediately following each data burst. The control word preceding a data burst indicates whether the following data burst is the start of a new packet or a continuation of a previously partially transmitted packet, and also indicates the port address of the following data burst. The control word immediately following a data burst indicates if the data burst contained an end of packet.

[0013] The status bus STAT is used to convey flow control information to SPI-4.2 transmitter 110. In a standard SPI-4.2 implementation, the flow control information is related to receive buffers (not shown) associated with the SPI-4.2 receiver. Each receive buffer reports whether it is “starving,” “hungry,” or “satisfied,” depending on buffer fullness. The status bus STAT transmits receive buffer status as a two-bit flow control word, where 00 represents starving, 01 represents hungry, and 10 represents satisfied. STAT repeatedly transmits a definable structure known as a calendar, consisting of a sync word (11), followed by at least one flow control word for each port in a defined sequence, followed by a parity word. The flow control words are updated for each calendar cycle.

[0014] When the SPI-4.2 transmitter and receiver are configured for dynamic deskew operation, the transmitter periodically transmits (e.g., separated by a configurable number of seconds) a training sequence. The training sequence consists of at least one idle control word followed by one or more repetitions of a twenty word training pattern. Each repetition of the training pattern consists of ten repeated training control words followed by ten repeated training data words. The training control words are orthogonal to the training data words, such that each data lane transitions at the end of each set of training control words and each set of training data words. This pattern is illustrated in FIG. 2 for the CTL signal and four of the data lanes. The training control word used herein is 0x0fff, where 0x represents hexadecimal notation. Consequently, the training data word is 0xf000.

[0015] Note in FIG. 2 that each lane contains a transition every 10 clock transitions, but the transitions are skewed (the illustrated skew pattern is merely illustrative). The receiver is expected to compensate for the skew during each dynamic training cycle.

[0016] The described embodiments phase-lock to DCLK and produce a sample clock MDCLK at a multiple of DCLK, e.g., at 16 times the rate of DCLK in one embodiment. FIG. 3 illustrates an expanded view of DCLK in the vicinity of FIG. 2 transition 10, along with the data lane training word transitions from FIG. 2 and MDCLK. At this scale, the relative skews between the training control word-to-training data word transitions are more apparent. CTL is shown transitioning at MDCLK −4, which is the nominal transition time were no skew apparent. DAT15 is shown transitioning at MDCLK −6, DAT14 is shown transitioning at MDCLK 0, DAT1 is shown transitioning at MDCLK 3, and DAT0 is shown transitioning at MDCLK −2. In this scenario, the optimal sample times for CTL, DAT15, DAT14, DAT1, and DAT0 are respective MDCLK values 0, −2, 4, 7, and 2.

[0017] FIG. 4 shows a partial block diagram of SPI-4.2 receiver 400, including a PLL 410, differential receivers 420C, 420.15, 420.1, and 420.0, delay MUX arrays DMC, DM15, DM1, and DM0, and a deskew logic/MUX select 430. PLL 410 receives and locks to DCLK to produce MDCLK (at sixteen times the DCLK frequency) and a receive clock RCLK locked to DCLK (at twice the DCLK frequency). Each data lane connects to a differential receiver, i.e., the differential bus pair CTL+, CTL− is input to differential receiver 420C, the differential bus pair D15+, D15− is input to differential receiver 420.15, etc. The output of each of the 17 differential receivers is supplied to a corresponding delay MUX array. Each delay MUX array communicates with deskew logic/MUX select 430, as will be explained further below. Finally, deskew logic/MUX select 430 outputs the received deskewed signals RCTL, RD15, . . . , RD1, RD0 for use in other portions of receiver 400 (not shown).

[0018] FIG. 5 illustrates one embodiment of a delay MUX array DMx. The delay MUX array comprises a delay MUX 500 comprising 16 serial delay registers F0 to F15, a group of 24 serial shift registers Sy.0 to Sy.23 coupled to the output of each serial delay register Fy, and a multi-bit comparator Cy coupled to Sy.0 to Sy.23 (or a subset thereof). The operation of each element of delay MUX array DMx will be described in turn.

[0019] Delay MUX 500 receives a signal x, which is the output of one of the differential receivers 420.x shown in FIG. 4. Delay register F0 loads the current value of signal x on each MDCLK negative edge, and thus loads 16 consecutive samples of signal x for each RCLK period. Delay register F1 loads the previous value of register F0 at the same time, and so on, such that delay MUX 500 holds the last 16 MDCLK samples of signal x, one of which will be closest to the desired sample time midway between consecutive data setup times. MUX select 430 sends a four-bit MUX address to delay MUX 500, indicating which of the 16 delay register outputs it has selected as the output value Rx. Each delay MUX array is supplied with its own MUX address selected to appropriately deskew that data lane.

[0020] The output of each delay register Fy is supplied to the head of a serial shift register chain Sy.0 . . . Sy.23. Each shift register in chain Sy.0 . . . Sy.23 is clocked with RCLK, and thus samples Fy each sixteen MDCLK samples. At any time, Sy.0 . . . Sy.23 contains a “snapshot” of the content of delay register Fy at each of the last 24 RCLK transitions. Thus considering all groups of serial shift registers, sixteen contesting “snapshots” of 24 values each are held.

[0021] Comparators C0 . . . C15 each compare one of the snapshots to the appropriate bits of a training sequence. For instance, if x is the output of differential receiver 420.0, the training bit sequence can be selected as 111111100000000001111111, corresponding to DCLK transitions 3 to 26 in FIG. 2 for DAT0. As the “snapshots” step through the training sequence during a training period, each comparator looks for this bit sequence in its group of shift registers. When (and if) the bit sequence is found, the comparator that found the sequence asserts a comparison output signal to deskew logic 430.

[0022] Note that in general some shift register groups will not match the sequence because they sample too near (or on the wrong side of) the setup period, but in general some band of comparators will all signal a match. The deskew logic finds the range of comparators that indicated a match, and selects the mean of this range as the desired address for the MUX select signal. Thus continuing with the example above for DAT0, comparators C7 to C11 could all indicate a training sequence match, causing deskew logic 430 to select C9 (and therefore F9) as its preferred delay MUX output for RD0. A delay MUX address of 1001 would be supplied to delay MUX array DM0 to select the output of F9 for use as RD0.

[0023] The above-described mechanism can be used once to set the deskew pattern for the bit lanes, or can be used dynamically to periodically adjust for changing skew/jitter patterns.

[0024] Although embodiments of the present disclosure have been described in detail, those skilled in the art should understand that they may make various changes, substitutions and alterations herein without departing from the spirit and scope of the present disclosure. Accordingly, all such changes, substitutions and alterations are intended to be included within the scope of the present disclosure as defined in the claims. It should be noted that the names given to modules and components of the system in the detailed description and claims are merely used to identified the modules and components for the sake of clarity and brevity and should not be used to limit or define the functionality or capability thereof unless explicitly described herein.