US20050220235A1

US20050220235A1 - System and method for aligning internal transmit and receive clocks

Info

Publication number: US20050220235A1
Application number: US11/130,506
Authority: US
Inventors: Donald Stark; Jun Kim; Stefanos Sidiropoulos
Original assignee: Individual
Current assignee: Individual
Priority date: 2000-02-07
Filing date: 2005-05-16
Publication date: 2005-10-06
Also published as: US6987823B1

Abstract

A system includes a master device connected to one or more slave devices via a channel, the channel communicating an externally generated first system clock towards the master device. A delay locked loop circuit receives the first system clock and a second phase feedback signal as inputs and generates a transmit clock signal. A phase offset circuit receives the transmit system clock and generates a phase shifted version of the transmit clock signal as a second system clock. A first phase detector receives a receive system clock and the transmit system clock and generates a first phase feedback signal. A delay element receives the first system clock and the first phase feedback signal and generates a delayed first system clock. A second phase detector receives the delayed first system clock and the second system clock and generates the second phase feedback signal.

Description

RELATED APPLICATIONS

This application is a continuation of Ser. No. 09/499,025, filed Feb. 7, 2000, which is hereby incorporated by reference.

BACKGROUND OF THE INVENTION

The present invention relates to a system and method for aligning two or more clock domains. More particularly, the present invention relates to a system and method for aligning transmit and receive clocks in a bus system.
FIG. 1A conceptually illustrates a bus system. The bus system generally comprises a master 3 and one or more slave devices (2 a . . . 2 n) connected via a channel comprising a number of signal lines or buses. Typically, a bi-directional bus communicates data between master 3 and slave devices (2 a . . . 2 n). Control information is communicated via the same or via a separate bus (not shown). Data and/or control information are communicated in relation to one or more clock signals. Master 3 is associated with an application 1. Application 1 may take many forms including a microprocessor, a memory controller, a graphics controller, etc. Application 1 may incorporate master 3 or be separately implemented.
In the example, shown in FIG. 1A, an externally generated Clock-To-Master (CTM), or first system clock signal, travels through the slave devices towards the master. At the master, CTM is turned around to form a Clock-From-Master (CFM), or a second system clock signal, which travel backs through the slave devices in a direction away from the master. In contemporary bus systems, the master and/or the slave devices typically includes an interface circuit (not shown) which controls the data and control information signals communicated between the master and the slave devices.
The relationship between application 1 and master 3 is further illustrated in FIG. 1B. Master 3 typically includes one or more delay locked loop (DLL) circuit(s), or similar circuit(s), which generates a receive clock (rclk) and a transmit clock (tclk). Generally speaking, the receive clock (rclk) controls the receiver functions in master 3 and the transmit clock (tclk) controls the transmit or data output functions in master 3. Thus, rclk and tclk define separate clock domains. This concept is illustrated by the relationships between receiver 3 a, output driver 3 b, and DLL 4 of FIG. 1B.
The receive clock (rclk) in the master is normally aligned with the knowledge that data being sent from the slave devices is communicated in a known relationship to CTM, and that this relationship is maintained as both the data signals and CTM traverse the channel towards the master. In other words, the receive clock (rclk) is normally phase aligned in a known relationship to CTM. This relationship is designed to maximize the timing margin for sampling the data at master 3. In many contemporary bus systems, data is transmitted 90° ahead of its corresponding CTM edge. As illustrated in FIG. 2, this relationship requires that the receive clock (rclk) lag CTM by a period of time equal to the nominal setup time for the receiver (T_SETUP _— _IR).
To achieve the foregoing, DLL 4 may be used. FIG. 3 illustrates an exemplary clock recovery circuit yielding the desired relationship comprising DLL 4 and flip-flop circuits (5 a . . . 5 e). Use of the receiver in the master as a phase detector for the DLL circuit assures that rclk properly lags CTM by the period T_SETUP _— _IR.
Referring to again to FIG. 1B, the transmit clock (tclk) is aligned with the knowledge that data being sent from the master to the slave devices is communicated with a known relationship to CFM, and that this relationship is maintained as both the data and CFM traverse the channel away from the master. This relationship is designed to maximize the timing margin for sampling the data at the slave devices.
In contemporary bus systems, it is common for data to be communicated 90° ahead of the corresponding CFM edge. Since there is a known, finite delay for the data traversing the output drivers in the master (output driver delay, T_OD), achieving the desired data to tclk timing relationship requires that the transmit clock (tclk) be (90°+T_OD) ahead of the corresponding CFM edge. This relationship is illustrated in FIG. 4.
A clock recovery circuit yielding the desired tclk relationships is shown in FIG. 5. Within this exemplary circuit, DLL 6 is used to align the transmit clock (tclk) which is applied to output drivers 10 a, 10 b . . . 10 n. The feedback path uses a 90° block 9 and a dummy output driver circuit 8 to achieve the desired phase relationship. A Zero degree Phase Detector (ZPD) is used to compare the feedback signal to CFM and drive DLL 6.
In addition to rclk and tclk, master 3 typically generates a third reference signal, Synclk. Synclk is used to control data exchanges between application 1 and master 3. That is, Synclk provides a reference for data signals received from the application by the master and for data signal sent from the master to the application. As illustrated in FIG. 1B, some contemporary bus systems formed Synclk by a dividing down the receiver clock (rclk) in divider circuit 3 c. Thus, the timing relationships for signals being communicated between the master and the application are ultimately referenced to Synclk which in turn is a product of rclk.
Unfortunately, as suggested above, a great number of control and data signals in the master must necessarily be referenced to tclk instead of Synclk/rclk. The existence of separate tclk and rclk domains within a bus system creates a number of synchronization concerns. For example, data from the application to be transmitted by the master to one or more slave devices must first be received in the master. This application-to-master data transfer is done in accordance with Synclk. However, the data is transmitted from the master to the one or more slave devices in accordance with tclk. The transition of such data from the rclk domain to the tclk domain is accomplished by “holding” the data in the master for some defined period of time.
Following conventional theory, CFM and CTM are identical except for their propagation direction. Thus, rclk and tclk would be similarly related, but for the finite timing delays necessarily introduced by operation of the receiver and the output driver circuits.
Unfortunately, as described in greater detail below, the ideal relationship between rclk and tclk do not hold in practice. Rather, timing delays introduced by circuit operations in varying voltage and temperature condition tend to skew the phase relationship between rclk and tclk. Recognizing that the electrical circuits in issue here will vary in their response time across a range of process, operating, and environment conditions, bus system designers must necessarily expand the synchronizing “hold” time periods within the master for data to accurately transition between the rclk and the tclk domains.
The timing diagram of FIG. 6 illustrates a set of ideal phase relationships between the clock signals described above. Consistent with contemporary practice, CTM and CFM are shown as a single signal. The phase relationship of rclk is T_SETUP _— _IRbehind CTM/CFM. Edge transitions for Synclk are synchronous with rclk. The phase relation of tclk is (90°+T_OD) ahead of CTM/CFM. Thus, if the delay of a clock signal through the output driver is (90°−T_SETUP _— _IR), then rclk and tclk will be separated in phase by 180°. These relationships are considered ideal in the working example.
Ideal sampling points for data transmitted from the application to the master correspond to the rising edge of rclk, as indicated by letters a, b, c, and d in FIG. 6. In other words, the setup and hold requirements which the application must adhere to are referenced to these edges.
However, as practically implemented within contemporary bus systems, the actual sampling of this data occurs at the falling edges of tclk, as indicated by aa, bb, cc, and dd of FIG. 6. Where the ideal phase relationships of FIG. 6 exist, the setup and hold requirements within the master consist of merely the setup and hold time of a flip-flop circuit sampling the data shifted by the input receiver setup time. Unfortunately, the ideal phase relationships of FIG. 6 rarely exist within bus systems.
To summarize, the setup time requirement for the data can be described as:
T _SETUP _— _Tdata=(T _OD +T _SETUP _— _IR−90°)+T _SETUP _— _FF, and
the hold time requirement for the data can be described as:
T _HOLD _— _Tdata =T _HOLD _— _FF−(90°−T _OD−T_SETUP _— _IR),
where T_SETUP _— _FF/T_HOLD _— _FFare the setup and hold times for the flip-flops sampling the data signals. Further, T_SETUP _— _Tdata/T_HOLD _— _Tdataare ideally referenced from the rising edge of rclk, and the optimal value for T_ODis (90°−T_SETUP _— _IR).
In actual implementation, however, the output driver delay (T_OD) is seldom equal to (90°−T_SETUP _— _IR). In fact, the delay at the output drivers will vary with operating conditions such as voltage and temperature. As a result, the ideal phase relations shown in FIG. 6 do not exist in practice. Recognizing this result, bus system designers have been forced to adopt rather loose standards for the sampling of data at the points indicated in FIG. 6. In other words, overall system timing requirements are squeezed by the necessity to accommodate a wide range of output driver delay times. In contemporary bus systems, the resulting timing restrictions are in the order of 3 ns for setup time and 2 ns for hold time. Such restrictions are a great burden on bus systems having rclk/tclk frequencies above several hundred MHz. This is particularly true since output driver delay times tends to decrease slower than the CTM cycle times.

SUMMARY OF THE INVENTION

A method of aligning clock signals in a bus system includes generating a transmit clock signal in a master, and arbitrarily adjusting the phase of the transmit clock signal while maintaining a first predefined phase relationship between the transmit clock signal and a second system clock. A further adjustment of the phase of the transmit clock signal may be made to have a second predefined phase relationship with a receive clock signal while maintaining the a first predefined phase relationship between the transmit clock signal and the second system clock. In one embodiment, the second predefined phase relationship between the transmit clock signal and the receive clock signal is 180°.
In another aspect, a method of aligning clock signals in a bus system includes generating a transmit clock signal in a master in relation to a first system clock, shifting the transmit clock signal phase by 90°, and passing the phase shifted transmit clock signal through an output driver circuit in the master to generate a second system clock. As a result and in contrast to the conventional expectation, the first and second system clocks need not be phase aligned.
In yet another aspect, the present invention provides a method of aligning system clocks in a bus system by generating a first system clock external to the master such that the first system clock propagates via the channel through the one or more slave towards the master, and generating in the master a second system clock having a phase relation to the first system clock defined such that, the phase difference between the first system clock and the second system clock is substantially equal to 90° minus the sum of the receiver setup delay and the output driver delay.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A is a diagram of a generalized bus system;
FIG. 1B is a diagram of the bus system of FIG. 1A in some additional detail;
FIG. 2 is a timing diagram illustrating an ideal phase relationship between CTM and rclk;
FIG. 3 is a diagram for an exemplary circuit nominally capable of implementing the timing relationship shown in FIG. 2;
FIG. 4 is a diagram illustrating an ideal phase relationship between CFM and tclk;
FIG. 5 is a diagram for an exemplary circuit nominally capable of implementing the timing relationship shown in FIG. 4;
FIG. 6 is a timing diagram summarizing the ideal phase relationships between CTM/CFM, rclk, Synclk, and tclk;
FIG. 7 is an exemplary circuit competent to provide a set of timing relationships in accordance with the present invention;
FIG. 8 is an alternative embodiment to the circuit shown in FIG. 7;
FIG. 9 is a timing diagram illustrating the resulting timing relationships of the present invention;
FIG. 10 conceptually illustrates the effect of channel length and slave device position along the channel to timing delay considerations inherent in the present invention;
FIG. 11 conceptually illustrates the requirement for cross clock domain synchronization within a bus system, assuming a memory system as a specific example;
FIG. 12 is a diagram of another exemplary synchronization circuit.

DETAILED DESCRIPTION

The maximum effective operating speed for a bus system is essentially the sum of critical path timing requirements. Further, data robustness in the bus system is a product of timing margins. Timing margins are impacted by a host of timing requirements. The restrictive setup and hold requirements explained above disadvantageously impact effective operating speed and timing margins.
The present invention addresses this problem by providing a system and method in which an ideal phase relationship between tclk and rclk domains can be maintained for all output driver delays across a range of bus system operating conditions. In one aspect, the present invention utilizes a CFM driver circuit which allows for arbitrary phase adjustments of tclk while maintaining the correct phase relationship between tclk and CFM, i.e., tclk being (90°+T_OD) ahead of CFM. Thereafter, the phase of tclk may be further adjusted until it has an optimal phase relationship with rclk, i.e. tclk being separated from rclk by 180°.
The circuit shown in FIG. 7 provides these desired phase relationships. In FIG. 7, CTM and the output of zero phase detector (ZPD) 26 are received in DLL circuit 20. The output of DLL 20 passes through 90° block 21 and buffer 22 a to be output at driver 23 as CFM. That is, 90° block 21 generates a signal tclk 90° which is delayed 90° from tclk. The signal tclk 90° is then used to generate the CFM signal through a standard output driver. The sum delay from these two blocks equals 90° plus the output driver delay (T_OD).
The output of DLL 20 also passes through buffer 22 b to yield tclk which is applied to the data output drivers 24 a, 24 b, . . . 24 n corresponding to Data 0, Data 1 . . . Data n. Along with rclk, the complement of tclk is applied to ZPD 26.
The circuit shown in FIG. 7 thus generates a tclk signal ahead of CFM by (90°+T_OD). Since tclk is used to generate data signals on the channel (Data 0, Data 1 . . . Data N), this relationship ensures that the data is 90° ahead of CFM, thereby maximizing data margins. Finally, the circuit maintains the optimal 180° relationship between rclk and tclk.
An alternative circuit is shown in FIG. 8. The alternative circuit substitutes a flip-flop circuit 27 for ZPD 26. Flip-flop 27 receives CTM as an input and the complement of tclk as a gating clock signal.
The exemplary circuits shown above may be modified to operate by using the complement of rclk, rather than tclk to control the output drivers. Since the feedback loop in the circuits above aligns tclk to the complement rclk, either signal may be used to control the transmit circuitry. Where the complement of rclk is used as the controlling signal, tclk exists merely to produce CFM.
All of these techniques yield the clock relationships shown in FIG. 9. Of note, the phase relationship between CTM and CFM is now different as compared with the conventional phase relationship normally assigned to CTM and CFM. The phase relationship between CTM and CFM may now be expressed as:
CTM−CFM=90°−(T _OD+T_SETUP _— _IR),
where T_ODequals the output driver delay and T_SETUP _— _IRequals the input receiver setup time. Thus, if T_OD+T_SETUP _— _IR>90°, then CFM trails CTM. If T_OD+T_SETUP _— _IR<90°, then CFM leads CTM.
With these desired relationships established, the application of the related clock signals to the devices in the bus system will now be examined. As can be understood from reference to system configuration illustrated in FIG. 1A, the phase relationship between CTM and CFM as defined by the present invention is different at each slave device depending on its position along the channel. Thus, individual slave devices must contain a mechanism making allowance for this arbitrary phase relationship.
FIG. 10 schematically illustrates this phenomenon. The delay between CFM and CTM at each slave device along the channel can be expressed as:
Total Delay=Intrinsic Delay+Cycle Delay+Fractional Delay.
Intrinsic delay is the time required to decode and execute an instruction at a slave device and does not vary between slave devices. For example, where the bus system is a memory system, intrinsic delay is the time required to decode an incoming “Read” request packet and fetch the desired data from memory.
Fractional delay is the extra delay that a slave device adds to the intrinsic delay such that the output of the desired data will be correctly aligned to the transmit clock (CTM). This delay linearly varies from zero when a slave device is near the upper end of a CTM/CFM cycle boundary to one cycle when a slave device is near the lower end of a CTM/CFM cycle boundary. As the CTM/CFM skew passes through a cycle boundary, the fractional delay value is reset to zero.
In the example illustrated in FIG. 10, five different cycle delay intervals are illustrated. However, a bus system may have any reasonable number of cycle delay intervals in accordance with its channel length, propagation speed, etc. No matter the actual size and configuration of the bus system, in order to maximize system bandwidth and minimize data bubbles on the channel, the master wants the apparent delay for each slave device to be constant. If the delay for each slave device consisted of only the intrinsic delay plus and the fractional delay, the master would “see” five different delays. For the example given in FIG. 10, this variable delay would range from zero to five for memory devices depending on the round trip distance on the bus between the master and each slave device. To avoid this problem, each slave device contains a programmable register which holds a cycle delay value corresponding to the number of additional cycles of delay added for each slave device. Again, with reference to the given example, the closest slave devices have an additional four clock cycles added by way of the register value. In contrast, the slave devices located farthest from the master have zero cycles of additional delay added. In this manner, each slave device presents the same apparent delay to the master.
A detailed circuit capable of introducing the fractional delay noted above has previously been described in commonly assigned U.S. Pat. No. 6,473,439, the subject matter of which is incorporated herein by reference. Whatever circuit actually used to achieve the desired results above, the concept of cross clock domain transition (i.e., fractional delay adjustment between receive and transmit clock domains) is illustrated in FIG. 11. FIG. 11 assumes a memory system as a working example of the bus system described throughout.
In FIG. 11, two delay locked loops (DLLs) are used to track the incoming clock signals. That is, CFM is applied to receiver DLL 30 and CTM is applied to transmit DLL 35. By tracking both CFM and CTM, the circuit ensures that control information and data being sent from the master to the slave device are received (and stored) at the appropriate times and that data being sent from the slave device to the master is transmitted at the appropriate time. Data transmitted from the master to the slave device is conceptually separated from associated control information in blocks 31 and 32. Data transmission circuitry for sending data from memory core 33 in the slave device to the master is indicated by 36.
Since CTM and CFM can have any phase relationship, care must be taken when passing data from the received clock domain (indicated by the dotted line in FIG. 11) to the transmit clock domain. A clock domain transition circuit 34 performs this cross domain handoff.
In one preferred embodiment, the clock domain transition circuit 34 chooses between two different delay paths based on the relative phases of CTM and CFM, such that setup and hold requirements in the transmits data block 36 are not violated. The transitions between these two delay paths occur at the CTM/CFM phase intervals of n*tcycle and (n+0.5)*tcycle. The first of these transitions causes the fractional delay to reset from one to zero. The second transition is required for correct circuit operation, but is not externally visible.
In conventional bus systems, the phase difference between CTM and CFM at a given slave device did not change appreciably. Rather, it was fixed by the length of the trace between the master and the slave device, as well as the propagation delay through the master. Accordingly, conventional bus systems would only activate the “Self Transition” function once during system initialization. During Self Transition the correct fractional delay would be determined, and based on an observation of received data at the master, for example, the cycle delay register would be programmed, such that each slave device presented the same apparent delay.
In contrast, the CTM and CFM phase difference resulting from application of the concepts of the present invention will vary according to operating conditions, i.e., changes in T_ODas a result of temperature, voltage etc. Thus, slave devices must be able to compensate for the changing phase relationship. There are a number of techniques which competently address this new requirement.
In a first technique, each slave device recalculates its fractional delay with sufficient frequency to effectively compensate for any variation in T_OD. This technique works well for bus systems whose total round trip is less than one cycle, because the update will require little controller overhead. However, systems exhibiting delays greater than one cycle are problematic because the apparent delay for slave devices near n*tcycle boundaries may change as the CFM to CTM phase relationship shifts. To compensate for this effect, the master would necessarily measure the delay for data arriving from each slave device following fractional delay adjustment, and reprogram, as necessary, the cycle delay register to maintain a constant apparent delay. Unfortunately, the overhead required to dynamically adjust both fractional and cycle delay components in this manner is prohibitive for many bus system applications.
Thus, in a preferred approach to this cycle boundary crossing problem, the slave device detects when it crosses a cycle delay boundary, and increments or decrements the cycle delay value in the cycle delay register accordingly. Such detection may be accomplished by noting when the fractional delay value goes back and forth across the 0 and 1 boundary.
In a second technique, sufficient margin is provided in the slave device CTM/CFM phase calibration circuitry to handle the T_ODvariation. Contemporary fractional delay circuits can automatically track up to 0.1*tcycles of CFM to CTM variation following operation of the Set Transition function. Further, variations in T_ODmay be significantly reduced by isolating the master (or master interface circuit) from environmental factors such as temperature and voltage.
A third technique is illustrated in FIG. 12. Within the exemplary circuit shown in FIG. 12, DLL 40 tries to align (CTM+delay) to CFM. During initial calibration, the delay amount is adjusted until rclk and tclk are 180° apart, i.e., their optimal phase relationship. Then the delay amount is held steady during an entire period of operation. The DLL will then maintain the relationship of (CTM+delay)=CFM, and will account for any variation in T_ODby adjusting the phase of tclk. In effect, this technique shifts the timing problem due to T_ODfrom CFM to tclk, thereby no longer ensuring the ideal relationship between tclk and rclk. However, tclk/rclk synchronization issues will be limited only to temperature and voltage variations since process variations may be compensated during the initial delay calibration. Re-calibration may be performed on the basis of shifts in temperature and voltage.
More specifically, CTM is applied to DLL 40 and delay line 46. The output of DLL 40 is applied to 90° block 41 and output through buffer 42 b as tclk. The output of 90° block 41 passes through buffer 42 a as tclk 90° and an output driver circuit 43 as CFM. A first zero phase detector circuit 45 receives rclk and the complement of tclk as inputs and also drives delay line 46. The output of delay line 46 and CFM are input to a second ZPD 47 which drives DLL 40.

Claims

1. A method of aligning clock signals in a system comprising a master and one or more slave devices connected via a channel, the system further comprising a first system clock propagating towards the master and a second system clock propagating away from the master, and the master further comprising a transmitter and a receiver, the method comprising:

generating a transmit clock signal in the master, wherein data is driven onto the channel in relation to the transmit clock signal; and

adjusting the phase of the transmit clock signal in the master while maintaining a first predetermined phase relationship between the transmit clock signal and the second system clock.

2. A method of aligning clock signals in a system comprising a master and one or more slave devices connected via a channel, the system further comprising a first system clock propagating towards the master and a second system clock propagating away from the master, the method comprising:

generating a transmit clock signal in the master in relation to the first system clock;

shifting a phase of the transmit clock signal phase by a predefined offset; and

passing the phase shifted transmit clock signal through an output driver circuit in the master to generate the second system clock, such that the first and second system clocks are not phase aligned.