CA2074008A1 - Quasi-synchronous information transfer and phase alignment means for enabling same - Google Patents
Quasi-synchronous information transfer and phase alignment means for enabling sameInfo
- Publication number
- CA2074008A1 CA2074008A1 CA002074008A CA2074008A CA2074008A1 CA 2074008 A1 CA2074008 A1 CA 2074008A1 CA 002074008 A CA002074008 A CA 002074008A CA 2074008 A CA2074008 A CA 2074008A CA 2074008 A1 CA2074008 A1 CA 2074008A1
- Authority
- CA
- Canada
- Prior art keywords
- unit
- clock
- data
- signals
- master unit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
- 238000012546 transfer Methods 0.000 title abstract description 20
- 230000001934 delay Effects 0.000 claims abstract description 14
- 230000010363 phase shift Effects 0.000 claims abstract description 12
- 239000000872 buffer Substances 0.000 claims description 50
- 230000002457 bidirectional effect Effects 0.000 claims description 18
- 230000001360 synchronised effect Effects 0.000 claims description 18
- 238000000034 method Methods 0.000 claims description 14
- 230000002265 prevention Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 210000004027 cell Anatomy 0.000 claims 3
- 210000000352 storage cell Anatomy 0.000 claims 2
- 238000010586 diagram Methods 0.000 description 12
- 238000013461 design Methods 0.000 description 4
- 240000008168 Ficus benjamina Species 0.000 description 1
- 239000003637 basic solution Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F13/00—Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
- G06F13/38—Information transfer, e.g. on bus
- G06F13/42—Bus transfer protocol, e.g. handshake; Synchronisation
- G06F13/4247—Bus transfer protocol, e.g. handshake; Synchronisation on a daisy chain bus
- G06F13/4256—Bus transfer protocol, e.g. handshake; Synchronisation on a daisy chain bus using a clocked protocol
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L7/00—Arrangements for synchronising receiver with transmitter
- H04L7/0016—Arrangements for synchronising receiver with transmitter correction of synchronization errors
- H04L7/005—Correction by an elastic buffer
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L7/00—Arrangements for synchronising receiver with transmitter
- H04L7/0008—Synchronisation information channels, e.g. clock distribution lines
Abstract
QUASI-SYNCHRONOUS INFORMATION TRANSFER AND PHASE
ALIGNMENT MEANS FOR ENABLING SAME
ABSTRACT
A quasi-synchronous information (clocking, data, control signals) transfer between a master unit A and at least one neighbor unit B offers the advantage that one transfer can occur every cycle where any interface delays are tolerated. When the master unit A sends its internal clock along with data and/or control signals to its neighbor unit B, the last named unit receives this clock and derives any and all locally required clocks from this clock. This keeps unit B exactly at the same frequency as unit A, although a phase shift occurs. Unit B also sends its internal clock along with data and/or control signals to unit A. When the clock signals arrive back at unit A, they will have exactly the same frequency as the internal clock of unit A but also an additional phase shift.
Compensation of the overall phase shift will be done by a phase alignment means to which all signals are sent on their way to unit A.
ALIGNMENT MEANS FOR ENABLING SAME
ABSTRACT
A quasi-synchronous information (clocking, data, control signals) transfer between a master unit A and at least one neighbor unit B offers the advantage that one transfer can occur every cycle where any interface delays are tolerated. When the master unit A sends its internal clock along with data and/or control signals to its neighbor unit B, the last named unit receives this clock and derives any and all locally required clocks from this clock. This keeps unit B exactly at the same frequency as unit A, although a phase shift occurs. Unit B also sends its internal clock along with data and/or control signals to unit A. When the clock signals arrive back at unit A, they will have exactly the same frequency as the internal clock of unit A but also an additional phase shift.
Compensation of the overall phase shift will be done by a phase alignment means to which all signals are sent on their way to unit A.
Description
D E S C R I P T I O N
QUASI-SYNCHRONOUS INFORMATION TRANSFER A~ PHASE
ALIGNMENT MEANS FOR ENABLING SAME
The invention generally refers to a method of quasi-synchronously transferring clocking-, data~, and/or control signals between two or more data processing units and to a phase alignment means for enabling same.
Figs. lA - C show a block diagram and timing diagrams of how a synchronous and an asynchronous interfaGe work.
Since they are widely known, only the important summary points are dealt with in the following:
Synchronous Interface (Fig. lA, B) - The clock signals of units A and B have the same frequency and the same phase relationship.
- One data transfer can occur every cycle via connecting cables 1 and/or 2.
- Any interface delays must be adjusted to exactly align the internal clocks of both units. For a system involving a cable and interface drivers, this may be a very difficult job to realize.
As can be seen from Fig. lB, a synchronous in-terface can be used to send data from unit A to unit B or vice versa without waiting to see if unit B or A received each data transfer successfully (correctly~. The synchronous interface, howsver, requires that the sender and receiver be in a "lock-step" with one another which means that the receiver reads the data at the same time the sender sends the data.
Fig. lB shows an example, wher0 a section of six clock cycles lA ... 6A, lB ... 6B respectively are used to transfer six data units (bits, bytes, words, etc.
':
: ' ' ~r~ 3 depending on the number of parallel lines) Dl ... D6, D10 ... D15 respectively from unit A to uni-t B and/or vice versa.
The clock pulses of the internal clock of unit A and unit B have the same frequency and phase, as can be seen from Fig. lB. Data unit Dl which is sent during clock cycle lA
of unit s A in-ternal clock via cable 1 will be received at unit B during the next clock cycle 2B of unit's B
internal clock. Data unit D10 which is sent from unit B
over cable 2 at the same time as data unit Dl will be received at unit A during clock cycle 2A of unit s A
internal clock.
Asynchronous Interface (Fig. lA, ~) - The clock signals of units A and B have different frequencies and different phase relationships.
- Multiple cycles are required for every data transfer.
- Any interface delays can exist, within the required limits of propagation delay and skew.
For this reason, many external interfaces from box to box are done asynchrono~asly.
Using an asynchronous interface, ~nit A sends data, one transfer at a time and waits to send the next transfer until unit B replies that it received the data correctly.
As can be seen from the pulse diagram in lines four and five of Fig. lC, two more interface lines are necessary for transmitting control signals. In the case of a data transfer from unit A to unit B a control signal from unit A is sent to unit B, meaning new data is on the bus. The other control signal is a reply signal from unit B sent to unit A, meaning unit B read the data.
GE9-91-010 3 ,~ , As further can be seen from lines three and eight of the pulse diagram in Fig. lC, multiple cyc].es are required for every data transfer. To transfer a data unit Dl from unit A to unit B almost four clock A cycles are necessary until this data unit Dl is completely captured at unit B.
When data from unit B have to be transferred asynchronously to unit A then the compleke process has to be reversed.
As can be seen from the description above, both interfaces have their advantages and disadvantages. The advantage of a synchronous interface is, that one data transfer can occur every cycle. Its disadvantage, however, is that any interface delays must be adjusted to exactly align the internal clocks of both units. An asynchronous interface has the advantage, that any interface delays can exist, within the required limits of propagation delay and sXew; its disadvantage, however, is that multiple cycles are required for any data transfer.
Therefore, it is the object of the present invention to avoid the disadvantages listed above of both interfaces and keep at the same time their advantages.
This problem is solved in accorclance with the invention by the featùres specified in the main claim.
Advantageous embodiments and fùrther developments of the subject matter of the invention may be seen from the subclaims.
Thus, the invention offers the data streaming advantage of synchronous interfaces without having to pay the penalty of asynchronous interfaces. The penalty that has to be paid is an added delay in the time it takes the first message from the sender to arrive at the receiver, but not in the overall data throughput.
E'urthermore, the invention allows various parallel processor system designs where various lengths of ~ 5 interconnecting lines or cables can be used in a highly flexible manner.
Embodiments of the invention will be described in detail below with reference to the accompanying drawings in whi.ch:
Fig. lA to lC is a schematic representation of known synchronous or asynchronous interfaces and their respective pulse diagrams, Fig. 2A, B is a schematic representation of a "quasi"
synchronous interface and its pulse diagram, Fig. 3A is a block diagram of an interface shown in Fig.
QUASI-SYNCHRONOUS INFORMATION TRANSFER A~ PHASE
ALIGNMENT MEANS FOR ENABLING SAME
The invention generally refers to a method of quasi-synchronously transferring clocking-, data~, and/or control signals between two or more data processing units and to a phase alignment means for enabling same.
Figs. lA - C show a block diagram and timing diagrams of how a synchronous and an asynchronous interfaGe work.
Since they are widely known, only the important summary points are dealt with in the following:
Synchronous Interface (Fig. lA, B) - The clock signals of units A and B have the same frequency and the same phase relationship.
- One data transfer can occur every cycle via connecting cables 1 and/or 2.
- Any interface delays must be adjusted to exactly align the internal clocks of both units. For a system involving a cable and interface drivers, this may be a very difficult job to realize.
As can be seen from Fig. lB, a synchronous in-terface can be used to send data from unit A to unit B or vice versa without waiting to see if unit B or A received each data transfer successfully (correctly~. The synchronous interface, howsver, requires that the sender and receiver be in a "lock-step" with one another which means that the receiver reads the data at the same time the sender sends the data.
Fig. lB shows an example, wher0 a section of six clock cycles lA ... 6A, lB ... 6B respectively are used to transfer six data units (bits, bytes, words, etc.
':
: ' ' ~r~ 3 depending on the number of parallel lines) Dl ... D6, D10 ... D15 respectively from unit A to uni-t B and/or vice versa.
The clock pulses of the internal clock of unit A and unit B have the same frequency and phase, as can be seen from Fig. lB. Data unit Dl which is sent during clock cycle lA
of unit s A in-ternal clock via cable 1 will be received at unit B during the next clock cycle 2B of unit's B
internal clock. Data unit D10 which is sent from unit B
over cable 2 at the same time as data unit Dl will be received at unit A during clock cycle 2A of unit s A
internal clock.
Asynchronous Interface (Fig. lA, ~) - The clock signals of units A and B have different frequencies and different phase relationships.
- Multiple cycles are required for every data transfer.
- Any interface delays can exist, within the required limits of propagation delay and skew.
For this reason, many external interfaces from box to box are done asynchrono~asly.
Using an asynchronous interface, ~nit A sends data, one transfer at a time and waits to send the next transfer until unit B replies that it received the data correctly.
As can be seen from the pulse diagram in lines four and five of Fig. lC, two more interface lines are necessary for transmitting control signals. In the case of a data transfer from unit A to unit B a control signal from unit A is sent to unit B, meaning new data is on the bus. The other control signal is a reply signal from unit B sent to unit A, meaning unit B read the data.
GE9-91-010 3 ,~ , As further can be seen from lines three and eight of the pulse diagram in Fig. lC, multiple cyc].es are required for every data transfer. To transfer a data unit Dl from unit A to unit B almost four clock A cycles are necessary until this data unit Dl is completely captured at unit B.
When data from unit B have to be transferred asynchronously to unit A then the compleke process has to be reversed.
As can be seen from the description above, both interfaces have their advantages and disadvantages. The advantage of a synchronous interface is, that one data transfer can occur every cycle. Its disadvantage, however, is that any interface delays must be adjusted to exactly align the internal clocks of both units. An asynchronous interface has the advantage, that any interface delays can exist, within the required limits of propagation delay and sXew; its disadvantage, however, is that multiple cycles are required for any data transfer.
Therefore, it is the object of the present invention to avoid the disadvantages listed above of both interfaces and keep at the same time their advantages.
This problem is solved in accorclance with the invention by the featùres specified in the main claim.
Advantageous embodiments and fùrther developments of the subject matter of the invention may be seen from the subclaims.
Thus, the invention offers the data streaming advantage of synchronous interfaces without having to pay the penalty of asynchronous interfaces. The penalty that has to be paid is an added delay in the time it takes the first message from the sender to arrive at the receiver, but not in the overall data throughput.
E'urthermore, the invention allows various parallel processor system designs where various lengths of ~ 5 interconnecting lines or cables can be used in a highly flexible manner.
Embodiments of the invention will be described in detail below with reference to the accompanying drawings in whi.ch:
Fig. lA to lC is a schematic representation of known synchronous or asynchronous interfaces and their respective pulse diagrams, Fig. 2A, B is a schematic representation of a "quasi"
synchronous interface and its pulse diagram, Fig. 3A is a block diagram of an interface shown in Fig.
2~, Fig. 3B is a block diagram of a phase alignment circuitry necessary for use in "quasi" synchronous interfaces, Figs.4A to 4D are schematic representations of a phase alignment circuitry shown in Fig. 3B;
Fig. S is a block diagram of an arrangement showing recursive use of a "quasi" synchronous interface, Fig. 6 is a block diagram showing multiple use of a "~uasi" synchronous interface, and Fig. 7A, 7B are block diagrams of a "c~uasi" synchronous interface used for bi-directional da-ta and control signals transfer.
In contrast to synchronous and asynchronous prior art interfaces (Figs. lA - C), Eig. 2 shows how a "~uasi-synchronous" interface can be used to keep the advantages of both interface types (synchronous, asynchronous) and to remove their disaclvantages. The basic solution is centered around a method of keeping the same frequency for both units, and compensating for the ~ GE9-91-010 5 2 if n 7 f~
difference in phase using a phase alignment means.
Keeping the same fre~uency allows the units to make one data transfer per cycle, while permitting a phase difference allows them to remove the re~uirement of making interface delay adjustments to align the units' clocks.
To understand the quasi-synchronous i.nterface better, reference is made to Figs. 2A, B. The term "clock" is used to refer to either a single clock or a group of clocks, depending on the implementation of the logic chosen. If a group of clocks is used~ then the skew between the individual clocks of the group as they propagate throughout the system mus-t be taken into consideration. However, for the purposes of this example no skew problems of this type are covered since they do not relate to the arrangements described.
It is assumed that unit A sends its internal clock along with data and control signals on separate lines 1, la to unit B, data and/or control lines on cable 1 and the clock on line la (cf. lines 2,3; Fig. 2B). Unit B
receives this clock and derives any and all local clocks from this clock. This keeps unit B at the exact same fre~uency as unit A, although a phase shift does occur.
The propagation delay as well as the phase shift can be lumped into one item and are referred to as DELAY A->B
(cf. line 4; Fig. 2B). Since unit B s clock is synchronized ~ith the data coming from unit A, it can accept data at a rate of one transfer per cycle (cf. line 5; Fig. 2B).
In the opposite direction unit B sends its internal clock along with data and control signals back to unit A via line 2a and cable 2. When this clock arrives at unit A, it will have exactly the same fre~uency as the clock internal to unit A, however, it will have a phase shift DELAY B->A equal -to the phase shift introduced from unit A to unit B, plus the phase shift in-troduced when sending from unit B to unit A (c~. line 8; Fig. 2B). The entire d ~ ~
interface enters unit A through a phase alignment means 3 whose purpose is to compensate for the phase shift of the signals arriving at the phase alignment means 3 from unit B relative to the internal clock of unit A (data and control signals leave the phase alignment means 3 aligned with unit A s internal clock). The internal design of the phase ali~nment means 3 is not important at this point.
Many designs can fulfill the re~uirement above e.g. a phase alignment buffer. What is important, however, is that both data and control lines from unit B to unit A
are treated as part of the data path in the buffer since both are phase shifted and need to be realigned. Another important point is that all phase buffer clocks used on the unit B side of the buffer are derived from the clock received from unit B (there is no clock adjusting or matching involved).
In Figs. 2A and 3A the phase alignment means 3 are shown as boxes (circuit arrangements) which are contained in unit A. The phase alignment means 3, however, can also be separate boxes which are connected to the units A and B.
Fig. 3A shows that the phase alignment means 3 is split into parts 3a and 3b which are re]ated to different clock domains. Whereas part 3a is related to the internal clock A (internal clock on A-side o~ the phase alignment means), is part 3b related to internal clock B (internal clock on B-side of the phase alignment means).
The clock generation takes place by a clock generator 4 only in unit A. All clock pulses necessary to control unit A or unit B are derived from the (master) clock pulses of clock generator 4. If necessary the clocks are distributed by various clock distributors 8, 9, ll - 14 within the units A-E to the units internal circuitry.
All data AD and control signals AC le~ving part 3a of the phase alignment means 3 are aligned with the internal clock of unit A. Another signal IGN which is derived from the phase alignment means 3 is used to characterize ,f .f', signals leaving the phase alignment means as being invalid. This ignore signal IGN will be described in greater detail in connection with Figs. 3B and Figs.
4A-E.
The system described above can be defined as a ~uasi-synchronous interface which has the properties listed below:
- Units have the same frequency, but different phase.
- One data transfer can occur every cycle.
- Any interface delays can be tolerated.
- Additional logic in the form of a phase alignment buffer must be added to the first unit.
An important part of the quasi-synchronous interface is the phase alignment buffer. This section describes one possible implementation of the buffer, however, other implementations are possible and the choice is largely implementation dependent. Designs which vary from -this implementation but still meet the quasi-synchronous interface must be regarded as being within the scope of this description.
The buffer chosen to be described here is designed to meet the re~uirements of an on-chip buffer which has a minimlim latency and does not have an interruptible data stream. This means that once the system is initiali~ed and a data stream starts, it is not interrupted under normal circumstances even for one cycle. A defect in the system such as the cable from unit A to unit B becoming disconnected would be an error condition and not a normal circumstance.
Referring to Fig. 3B, it can be seen, that both the signals on the data 2 and control 2b lines from unit B to ~r~
the phase alignment buffer 3 are merged toge~her and treated as data and passed through an input selector 5 into a particular location I-N in the buffer array 6.
This location is defined by the contents of a WRITE
pointer stored in 15 which acts as an address pointer onto the locations of the buffer array. The WRITE pointer is updated (e.g.incremented) by adder 10 on every clock cycle which comes across the interface from unit B. Its value is passed to the other side 3a using metastability prevention compensation techniques (MS logic) 7 for comparison with a READ pointer. The WRITE pointer wraps back upon itself when it reaches the last address of the array (e.g. modulus 4 array; WRITE
0-1-2-3-0-1-2-3-0-...). The array has the property of being able to be written with one clock (from B) and read under control of a second different clock (from A).
On the read side 3a of the array, buffer locations I-N
are read on every clock cycle from the internal clock from unit A. Reading outputs the interface conditions ~ritten in by the B side with a given offset of time. The READ pointer stored in 17 chooses the array location to be gated onto the read side data and control lines which are fed into the receive logic of unit A. The READ
pointer also wraps around itself when it reaches the last address of the array, just like the WRITE pointer does.
Data AD and control AC lines leaving the phase buffer 3 via mul-tiplexer 18 and going to the receive logic of unit A are exactly synchronized both in frequency and phase to unit A s clock because they are read out of the array using unit A's internal clock. When both the READ and WRITE pointers reach the end of the buffer array 6, they wrap around to the first buffer location.
Metastability (MS) logic 7 and Read Control Logic 16 are used to determine whether the information stored in the currently indicated buffer is valid or not, and whether to advance the READ pointer or not. BufEer contents becomes invalid if the write side clock stops for any reason (controlled stop, error stop) while the read side clock is still running. This control logic resides on unit A s c:Lock domain. It ls broken into two parts. The first part 7 captures a valid copy of the WRITE pointer into the same clock domain as the READ pointer (on unit A s internal clockj. The second part 16 uses this copy of the WRITE pointer to decide the validity of the buffer contents and whether to increment the READ pointer. The WRITE pointer must always be ahead of the READ pointer.
If both point to the same buffer location, the read output becomes invalid and has to be ignored by the unit A receive logic.
For the moment, it is assumed that a valid copy of the WRITE pointer can be obtained on the read side clock domain; this copy is called WRITECOPY WC. The read control logic 16 is used to determine the da-ta validity and advancing of the READ pointer as follows:
IF ~READ = WRITECOPY) THEN * The pointers are the same ACTIVATE IGNORE, * Tell receive logic data is INVAL,ID
DO NOT INCREMENT READ * Wait until pointers are different ELSE
DEACTIVATE IGNORE, * Tell receive logic data is VALID
INCRE~ENT R~AD. ~ Point to the next array location With this read control logic 16, if the IGNORE line becomes active it means that the write side of the phase buffer did not receive a clock from unit B. This can happen in two situations. The first case is when the system is initializing and the clocks which have left unit A are still propagating through unit B and back into the phase buffer of unit A. Until the first clock reaches the phase buffer WRITE pointer, I~NORE will be active.
The second case is when the path from A to B to phase buffer 3 was broken somewhere. In either case the receive GE9-91-01~ 10 logic in unit A should treat bo-th the data and control lines leaving the phase buffer as invalid until the IGNORE line becomes inactive.
A detailed description of the buffer pointer synchronization and adjustment (by MS logic 17) will be given in connection with Fig. 4A - D.
Recursive use of the Quasi-Synchronous Interface:
The use of the quasi-synchronous interface is not limited to two units A, B. It can be expanded through any collection of units A, B, C, ... as shown in Fig. 5. As long as the same clock frequency is kept, any number of phase delays can be included before reaching the phase alignment buffers 3/1 ... 3/3.
The example configuration in Fig. 5 shows three possible quasi-synchronous paths:
1~ Unit A -> unit B -~ BUFFER in unit A (3/1) 2) Unit B -> unit C -> BUFFER in unit B (3/3) 3) Unit A -> unit B -> unit C -> BUFFER in unit A (3/2) Each phase alignment bu~fer 3/1 - 3/3 has a separate set of output lines on which the aligned data signals AD, the aligned control signals AC and the ignore signal are provided to their respective units A, B and/or C. Clock distributors 11, 1~ may also be used in those units where it is necessary to derive more clocks from the master clocl~.
Multiple Use of the Quasi-Synchronous Interface:
Fig. 6 shows that more than one interface can be controlled from a single unit A. There is no limitation on the number of .interfaces that can be controlled, however, one phase alignment buffer (3/1, 3/2, 3/4, 3/5) per returning interface is necessary.
GE9-91-010 11 ~ r( l~r ~ 3 ~3 Mapping a Bidirectional Interface into a Quasi-Synchronous Interface:
One important feature of the quasi-synchronous interface is that everything is broken into two unidirectional interfaces, one going each way. This is important to keep the phase shifts adding in the same direction until the phase buffer can realign the information. However, this does not mean that bidirectional interfaces cannot be used, they must only be handled specially.
Figs. 7A, B show how the bidirectional lines 20 must be split on unit A. Unit B sees no difference in its implementation of the bidirectional control logic.
However, unit A has a special switch function 23 built in around the normal phase alignment buffer 3.
When unit A is in receive mode (switch in position R), the data path 20 is gated such that data coming from the interface from unit B is routed through the phase buffer 3 before entering unit A's receive logic. When unit A is in send mode (switch in position S), the unit A s output path is connected directly to the interface to unit B.
Also during send mode, unit A's receive logic should be connected to a NOP value circuitry (NOP = no operation) that will indicate to unit A's receive logic that the interface is sending and not receiving. One can also leave unit A s receive logic connected to the phase buffer, but since it is still running one does not know what is coming out of it. If unit A's receive logic understands that it should discard the information coming into i-t, th~n no special value needs to be assigned.
When in switching modes, a number of NOP cycles should be performed by unit A's receive logic since the data path through the phase buffer must be reestablished with valid data before normal data processing can start. The number of NOP cycles is implementation dependant.
In the Figs. and the above description, data is used to refer to the signals which are bidirectional in nature and control to refer to the signals which are unidirectional. The terms data and control were chosen for the purposes of illustration as this is how a typical bidirectional interface is implemented. Bidirectional signals can also have control functions, likewise unidirectional signals can have data functions. Other implementations of handling bidirectional interfaces are possible; this is only one particular example used for illustration purposes.
Getting a copy of WRITE to the READ side clock domain:
Reference is made to Fig. 3B, 4A-E for the ~ollowing description of -the phase alignment buffer write and read control. For ease of drawing, the WRITE pointer is designated W.
To obtain a stable cop~ of the WRITE pointer W across the clock boundary on every clock cycle, a series of delays ~
is used to make four different copies F, G, H, J of the contents of the WRITE pointer. The value of ~ should be strictly chosen by the relationship A>DmaX, DmaX being a maximal drift, caused by changes in temperature, threshold values, edge slope and noise, interference, ground and voltage shifts etc.. Accordi~gly, b should be less than 1/3 clock cycle.
Fig.4A shows the WRITE poin-ter W on the interface clock domain IFD being sent across clock boundary CB to the internal clock domain IND of unit A and being captured in registers F,G,H, and J within MS-LOGIC 7 with a delay between F and G, G and H, H and J.
Fig.4B shows the relative positions of the copies o the contents of W in timing diagram format. As can be seen, data arriving at F is exactly aligned with the changes happening in W (not counting physical delays in the implementation of the circuit). Data arriving at G is delayed in time b~ 1 ~, at H by 2 ~, and at J by 3 ~. The IND clock will latch all our copies at one time, thus the contents of registers E,G,H,J represent the value of W at different times.
Fig.4C is the same picture redrawn showing the delays Q
at different capturing points for F,G,H,J even though in reality all four are being latched simultaneously. The relationship between F,G,~,J is fixed due to the delays ~
(J is the oldest copy of W, F is the newest) however the actual contents of each varies based on the alignment between unit A's internal clock and the interface clock.
This alignment is covered in the next figure which also brings in the concept of "drift".
Fig.4D varies the relationships between the read side clock and the write side clock by moving the point where the write side clock changes data in the WRITE pointer.
In the irst line, it is shown that the write clock occurs well before the read clock and all four copies of W are e~ual (J=H=G=F). ~owever, the point where the write clock occurs is shown as a bar with a center point, not just a point. This is due to what is defined as "drift"
for the purposes of this description.
The "drift" is the amount of chang~e which can be expected in the write clock edge, relative to the very first clock edge received by the phase alignment buffer. Over time, one would not expect the trigger point of the WRITE
pointer to remain the same because things such as temperature changes, threshold changes, clock edge slope changes, noise, interference, ground and voltage shifts, etc. can all affect thP point of WRITE pointer triggering. One would expect that this time is small, however, it is not negligible and the value of the delays ~ must be chosen to be larger than the maximum possible drift one can expect over the system lietime. The value of this drift is highly implementation dependant, however for this purposes it ls assumed that it is in the 2-3 nanosecond range. Therefore, the value chosen for the delay Q should be greater than 3 nanoseconds. The value must not be larger than 1/3 the cycle time, however, so that all four copies of the WRITE pointer are valid each cycle.
Returning to Fig. 4D, there are 9 possible alignments of the write side clock versus the read side clock. These are as follows.
1) W clock well before R clock J=H=G=F WRITEOPY=H
2) W clock such that J is unstabl~ J<>H=G=F WRITECOPY=F
3) W clock between J and H, J stable J<>H=G=F WRITECOPY=F
Fig. S is a block diagram of an arrangement showing recursive use of a "quasi" synchronous interface, Fig. 6 is a block diagram showing multiple use of a "~uasi" synchronous interface, and Fig. 7A, 7B are block diagrams of a "c~uasi" synchronous interface used for bi-directional da-ta and control signals transfer.
In contrast to synchronous and asynchronous prior art interfaces (Figs. lA - C), Eig. 2 shows how a "~uasi-synchronous" interface can be used to keep the advantages of both interface types (synchronous, asynchronous) and to remove their disaclvantages. The basic solution is centered around a method of keeping the same frequency for both units, and compensating for the ~ GE9-91-010 5 2 if n 7 f~
difference in phase using a phase alignment means.
Keeping the same fre~uency allows the units to make one data transfer per cycle, while permitting a phase difference allows them to remove the re~uirement of making interface delay adjustments to align the units' clocks.
To understand the quasi-synchronous i.nterface better, reference is made to Figs. 2A, B. The term "clock" is used to refer to either a single clock or a group of clocks, depending on the implementation of the logic chosen. If a group of clocks is used~ then the skew between the individual clocks of the group as they propagate throughout the system mus-t be taken into consideration. However, for the purposes of this example no skew problems of this type are covered since they do not relate to the arrangements described.
It is assumed that unit A sends its internal clock along with data and control signals on separate lines 1, la to unit B, data and/or control lines on cable 1 and the clock on line la (cf. lines 2,3; Fig. 2B). Unit B
receives this clock and derives any and all local clocks from this clock. This keeps unit B at the exact same fre~uency as unit A, although a phase shift does occur.
The propagation delay as well as the phase shift can be lumped into one item and are referred to as DELAY A->B
(cf. line 4; Fig. 2B). Since unit B s clock is synchronized ~ith the data coming from unit A, it can accept data at a rate of one transfer per cycle (cf. line 5; Fig. 2B).
In the opposite direction unit B sends its internal clock along with data and control signals back to unit A via line 2a and cable 2. When this clock arrives at unit A, it will have exactly the same fre~uency as the clock internal to unit A, however, it will have a phase shift DELAY B->A equal -to the phase shift introduced from unit A to unit B, plus the phase shift in-troduced when sending from unit B to unit A (c~. line 8; Fig. 2B). The entire d ~ ~
interface enters unit A through a phase alignment means 3 whose purpose is to compensate for the phase shift of the signals arriving at the phase alignment means 3 from unit B relative to the internal clock of unit A (data and control signals leave the phase alignment means 3 aligned with unit A s internal clock). The internal design of the phase ali~nment means 3 is not important at this point.
Many designs can fulfill the re~uirement above e.g. a phase alignment buffer. What is important, however, is that both data and control lines from unit B to unit A
are treated as part of the data path in the buffer since both are phase shifted and need to be realigned. Another important point is that all phase buffer clocks used on the unit B side of the buffer are derived from the clock received from unit B (there is no clock adjusting or matching involved).
In Figs. 2A and 3A the phase alignment means 3 are shown as boxes (circuit arrangements) which are contained in unit A. The phase alignment means 3, however, can also be separate boxes which are connected to the units A and B.
Fig. 3A shows that the phase alignment means 3 is split into parts 3a and 3b which are re]ated to different clock domains. Whereas part 3a is related to the internal clock A (internal clock on A-side o~ the phase alignment means), is part 3b related to internal clock B (internal clock on B-side of the phase alignment means).
The clock generation takes place by a clock generator 4 only in unit A. All clock pulses necessary to control unit A or unit B are derived from the (master) clock pulses of clock generator 4. If necessary the clocks are distributed by various clock distributors 8, 9, ll - 14 within the units A-E to the units internal circuitry.
All data AD and control signals AC le~ving part 3a of the phase alignment means 3 are aligned with the internal clock of unit A. Another signal IGN which is derived from the phase alignment means 3 is used to characterize ,f .f', signals leaving the phase alignment means as being invalid. This ignore signal IGN will be described in greater detail in connection with Figs. 3B and Figs.
4A-E.
The system described above can be defined as a ~uasi-synchronous interface which has the properties listed below:
- Units have the same frequency, but different phase.
- One data transfer can occur every cycle.
- Any interface delays can be tolerated.
- Additional logic in the form of a phase alignment buffer must be added to the first unit.
An important part of the quasi-synchronous interface is the phase alignment buffer. This section describes one possible implementation of the buffer, however, other implementations are possible and the choice is largely implementation dependent. Designs which vary from -this implementation but still meet the quasi-synchronous interface must be regarded as being within the scope of this description.
The buffer chosen to be described here is designed to meet the re~uirements of an on-chip buffer which has a minimlim latency and does not have an interruptible data stream. This means that once the system is initiali~ed and a data stream starts, it is not interrupted under normal circumstances even for one cycle. A defect in the system such as the cable from unit A to unit B becoming disconnected would be an error condition and not a normal circumstance.
Referring to Fig. 3B, it can be seen, that both the signals on the data 2 and control 2b lines from unit B to ~r~
the phase alignment buffer 3 are merged toge~her and treated as data and passed through an input selector 5 into a particular location I-N in the buffer array 6.
This location is defined by the contents of a WRITE
pointer stored in 15 which acts as an address pointer onto the locations of the buffer array. The WRITE pointer is updated (e.g.incremented) by adder 10 on every clock cycle which comes across the interface from unit B. Its value is passed to the other side 3a using metastability prevention compensation techniques (MS logic) 7 for comparison with a READ pointer. The WRITE pointer wraps back upon itself when it reaches the last address of the array (e.g. modulus 4 array; WRITE
0-1-2-3-0-1-2-3-0-...). The array has the property of being able to be written with one clock (from B) and read under control of a second different clock (from A).
On the read side 3a of the array, buffer locations I-N
are read on every clock cycle from the internal clock from unit A. Reading outputs the interface conditions ~ritten in by the B side with a given offset of time. The READ pointer stored in 17 chooses the array location to be gated onto the read side data and control lines which are fed into the receive logic of unit A. The READ
pointer also wraps around itself when it reaches the last address of the array, just like the WRITE pointer does.
Data AD and control AC lines leaving the phase buffer 3 via mul-tiplexer 18 and going to the receive logic of unit A are exactly synchronized both in frequency and phase to unit A s clock because they are read out of the array using unit A's internal clock. When both the READ and WRITE pointers reach the end of the buffer array 6, they wrap around to the first buffer location.
Metastability (MS) logic 7 and Read Control Logic 16 are used to determine whether the information stored in the currently indicated buffer is valid or not, and whether to advance the READ pointer or not. BufEer contents becomes invalid if the write side clock stops for any reason (controlled stop, error stop) while the read side clock is still running. This control logic resides on unit A s c:Lock domain. It ls broken into two parts. The first part 7 captures a valid copy of the WRITE pointer into the same clock domain as the READ pointer (on unit A s internal clockj. The second part 16 uses this copy of the WRITE pointer to decide the validity of the buffer contents and whether to increment the READ pointer. The WRITE pointer must always be ahead of the READ pointer.
If both point to the same buffer location, the read output becomes invalid and has to be ignored by the unit A receive logic.
For the moment, it is assumed that a valid copy of the WRITE pointer can be obtained on the read side clock domain; this copy is called WRITECOPY WC. The read control logic 16 is used to determine the da-ta validity and advancing of the READ pointer as follows:
IF ~READ = WRITECOPY) THEN * The pointers are the same ACTIVATE IGNORE, * Tell receive logic data is INVAL,ID
DO NOT INCREMENT READ * Wait until pointers are different ELSE
DEACTIVATE IGNORE, * Tell receive logic data is VALID
INCRE~ENT R~AD. ~ Point to the next array location With this read control logic 16, if the IGNORE line becomes active it means that the write side of the phase buffer did not receive a clock from unit B. This can happen in two situations. The first case is when the system is initializing and the clocks which have left unit A are still propagating through unit B and back into the phase buffer of unit A. Until the first clock reaches the phase buffer WRITE pointer, I~NORE will be active.
The second case is when the path from A to B to phase buffer 3 was broken somewhere. In either case the receive GE9-91-01~ 10 logic in unit A should treat bo-th the data and control lines leaving the phase buffer as invalid until the IGNORE line becomes inactive.
A detailed description of the buffer pointer synchronization and adjustment (by MS logic 17) will be given in connection with Fig. 4A - D.
Recursive use of the Quasi-Synchronous Interface:
The use of the quasi-synchronous interface is not limited to two units A, B. It can be expanded through any collection of units A, B, C, ... as shown in Fig. 5. As long as the same clock frequency is kept, any number of phase delays can be included before reaching the phase alignment buffers 3/1 ... 3/3.
The example configuration in Fig. 5 shows three possible quasi-synchronous paths:
1~ Unit A -> unit B -~ BUFFER in unit A (3/1) 2) Unit B -> unit C -> BUFFER in unit B (3/3) 3) Unit A -> unit B -> unit C -> BUFFER in unit A (3/2) Each phase alignment bu~fer 3/1 - 3/3 has a separate set of output lines on which the aligned data signals AD, the aligned control signals AC and the ignore signal are provided to their respective units A, B and/or C. Clock distributors 11, 1~ may also be used in those units where it is necessary to derive more clocks from the master clocl~.
Multiple Use of the Quasi-Synchronous Interface:
Fig. 6 shows that more than one interface can be controlled from a single unit A. There is no limitation on the number of .interfaces that can be controlled, however, one phase alignment buffer (3/1, 3/2, 3/4, 3/5) per returning interface is necessary.
GE9-91-010 11 ~ r( l~r ~ 3 ~3 Mapping a Bidirectional Interface into a Quasi-Synchronous Interface:
One important feature of the quasi-synchronous interface is that everything is broken into two unidirectional interfaces, one going each way. This is important to keep the phase shifts adding in the same direction until the phase buffer can realign the information. However, this does not mean that bidirectional interfaces cannot be used, they must only be handled specially.
Figs. 7A, B show how the bidirectional lines 20 must be split on unit A. Unit B sees no difference in its implementation of the bidirectional control logic.
However, unit A has a special switch function 23 built in around the normal phase alignment buffer 3.
When unit A is in receive mode (switch in position R), the data path 20 is gated such that data coming from the interface from unit B is routed through the phase buffer 3 before entering unit A's receive logic. When unit A is in send mode (switch in position S), the unit A s output path is connected directly to the interface to unit B.
Also during send mode, unit A's receive logic should be connected to a NOP value circuitry (NOP = no operation) that will indicate to unit A's receive logic that the interface is sending and not receiving. One can also leave unit A s receive logic connected to the phase buffer, but since it is still running one does not know what is coming out of it. If unit A's receive logic understands that it should discard the information coming into i-t, th~n no special value needs to be assigned.
When in switching modes, a number of NOP cycles should be performed by unit A's receive logic since the data path through the phase buffer must be reestablished with valid data before normal data processing can start. The number of NOP cycles is implementation dependant.
In the Figs. and the above description, data is used to refer to the signals which are bidirectional in nature and control to refer to the signals which are unidirectional. The terms data and control were chosen for the purposes of illustration as this is how a typical bidirectional interface is implemented. Bidirectional signals can also have control functions, likewise unidirectional signals can have data functions. Other implementations of handling bidirectional interfaces are possible; this is only one particular example used for illustration purposes.
Getting a copy of WRITE to the READ side clock domain:
Reference is made to Fig. 3B, 4A-E for the ~ollowing description of -the phase alignment buffer write and read control. For ease of drawing, the WRITE pointer is designated W.
To obtain a stable cop~ of the WRITE pointer W across the clock boundary on every clock cycle, a series of delays ~
is used to make four different copies F, G, H, J of the contents of the WRITE pointer. The value of ~ should be strictly chosen by the relationship A>DmaX, DmaX being a maximal drift, caused by changes in temperature, threshold values, edge slope and noise, interference, ground and voltage shifts etc.. Accordi~gly, b should be less than 1/3 clock cycle.
Fig.4A shows the WRITE poin-ter W on the interface clock domain IFD being sent across clock boundary CB to the internal clock domain IND of unit A and being captured in registers F,G,H, and J within MS-LOGIC 7 with a delay between F and G, G and H, H and J.
Fig.4B shows the relative positions of the copies o the contents of W in timing diagram format. As can be seen, data arriving at F is exactly aligned with the changes happening in W (not counting physical delays in the implementation of the circuit). Data arriving at G is delayed in time b~ 1 ~, at H by 2 ~, and at J by 3 ~. The IND clock will latch all our copies at one time, thus the contents of registers E,G,H,J represent the value of W at different times.
Fig.4C is the same picture redrawn showing the delays Q
at different capturing points for F,G,H,J even though in reality all four are being latched simultaneously. The relationship between F,G,~,J is fixed due to the delays ~
(J is the oldest copy of W, F is the newest) however the actual contents of each varies based on the alignment between unit A's internal clock and the interface clock.
This alignment is covered in the next figure which also brings in the concept of "drift".
Fig.4D varies the relationships between the read side clock and the write side clock by moving the point where the write side clock changes data in the WRITE pointer.
In the irst line, it is shown that the write clock occurs well before the read clock and all four copies of W are e~ual (J=H=G=F). ~owever, the point where the write clock occurs is shown as a bar with a center point, not just a point. This is due to what is defined as "drift"
for the purposes of this description.
The "drift" is the amount of chang~e which can be expected in the write clock edge, relative to the very first clock edge received by the phase alignment buffer. Over time, one would not expect the trigger point of the WRITE
pointer to remain the same because things such as temperature changes, threshold changes, clock edge slope changes, noise, interference, ground and voltage shifts, etc. can all affect thP point of WRITE pointer triggering. One would expect that this time is small, however, it is not negligible and the value of the delays ~ must be chosen to be larger than the maximum possible drift one can expect over the system lietime. The value of this drift is highly implementation dependant, however for this purposes it ls assumed that it is in the 2-3 nanosecond range. Therefore, the value chosen for the delay Q should be greater than 3 nanoseconds. The value must not be larger than 1/3 the cycle time, however, so that all four copies of the WRITE pointer are valid each cycle.
Returning to Fig. 4D, there are 9 possible alignments of the write side clock versus the read side clock. These are as follows.
1) W clock well before R clock J=H=G=F WRITEOPY=H
2) W clock such that J is unstabl~ J<>H=G=F WRITECOPY=F
3) W clock between J and H, J stable J<>H=G=F WRITECOPY=F
4) W cloc~ such that H is unstable J<>H<>G=F WRITECOPY=F
5) W clock between H and G, H ~table J=H<>G=F WRITECOPY=F
6) W clock such that G is unstable J=H<>G<>F ~RITECOPY=F
7) W clock between G and F, G stable J=H=G<>F WRITECOPY=H
8) W clock such that F is unstable J=H-G<>F WRITECOPY=H
9) W clock well after R cloc~ J=H=G=F WRITECOPY~H
Since the delays between F,G,H,J are larger than the maximum drift range, we know that once the relationship between write side clock and read side clock is known, the relationship between the values of F,G,H,J will also be known and be stable. Therefore, we only need to sample F,G,H,J once after a WRITE clock has happened to know which copy of the WRITE pointer to use as WRITECOPY from then on.
A logic function is implemented in Fig. 3B on the read side clock domain to provide a stable copy of the WRITE
pointer to the read side clock domain. The reduced function is as follows.
During initialization, R=F=G=H=J.
DO WHILE (READ ~> F) WRITECOPY = F
END
GE9-91-010 15 ~j; rj~ 5~
WRITE has moved when READ<>F and the above loop is ex:ited.
IF (J=H=G) THEN
WRITECOPY = H * Unstable zone is not between J and G
ELSE
WRITECOPY = F * Unstable zone is somewhere between J
and G
rom now on, WRITECOPY always comes from this copy of W.
Since the delays between F,G,H,J are larger than the maximum drift range, we know that once the relationship between write side clock and read side clock is known, the relationship between the values of F,G,H,J will also be known and be stable. Therefore, we only need to sample F,G,H,J once after a WRITE clock has happened to know which copy of the WRITE pointer to use as WRITECOPY from then on.
A logic function is implemented in Fig. 3B on the read side clock domain to provide a stable copy of the WRITE
pointer to the read side clock domain. The reduced function is as follows.
During initialization, R=F=G=H=J.
DO WHILE (READ ~> F) WRITECOPY = F
END
GE9-91-010 15 ~j; rj~ 5~
WRITE has moved when READ<>F and the above loop is ex:ited.
IF (J=H=G) THEN
WRITECOPY = H * Unstable zone is not between J and G
ELSE
WRITECOPY = F * Unstable zone is somewhere between J
and G
rom now on, WRITECOPY always comes from this copy of W.
Claims (11)
1. Method of quasi-synchronously transferring clocking, data, and/or control signals between two or more data processing units (A,B,..., Fig.2A) characterized by performing the following steps:
generating one or more clock pulses on one single master unit (e.g. A), passing the clock pulses (internal clock A) together with the data and/or control pulses having the same frequency and phase as said clock pulses from said master unit over a unidirectional synchronous channel (1,1a) to a neighbor unit (e.g. B) where the transferred clock pulses define the local clock(s) (e.g. internal clock B), passing back the clocking, data, and/or control signals over a unidirectional channel (2,2a) from said neighbor unit to said master unit where said signals occur with the same frequency but with a shift in phase and compensating said phase shift by a phase alignment means (3) through which said unidirectional channel (2,2a) is connected to said master unit and where said phase alignment means adjusts the data, and/or control signals received from the neighbor unit to exactly the phase of the sending master unit, which signals forming the aligned data and/or control signals (AD, AC) for further usage.
generating one or more clock pulses on one single master unit (e.g. A), passing the clock pulses (internal clock A) together with the data and/or control pulses having the same frequency and phase as said clock pulses from said master unit over a unidirectional synchronous channel (1,1a) to a neighbor unit (e.g. B) where the transferred clock pulses define the local clock(s) (e.g. internal clock B), passing back the clocking, data, and/or control signals over a unidirectional channel (2,2a) from said neighbor unit to said master unit where said signals occur with the same frequency but with a shift in phase and compensating said phase shift by a phase alignment means (3) through which said unidirectional channel (2,2a) is connected to said master unit and where said phase alignment means adjusts the data, and/or control signals received from the neighbor unit to exactly the phase of the sending master unit, which signals forming the aligned data and/or control signals (AD, AC) for further usage.
2. Method of quasi-synchronously transferring clocking, data, and/or control signals between two or more data processing units (A,B,...; Fig.2A) characterized by performing the following steps:
generating one or more clock pulses on one single master unit (e.g. A), passing the clock pulses (internal clock A) -together with the control pulses over a unidirectional synchronous channel (1,1a) and with data pulses over a bidirectional channel (20), said pulses having the same frequency and phase as said clock pulses from said master unit to a neighbor unit (e.g. B) where the transferred clock pulses define the local clock(s) (e.g. internal clock B), passing back the clocking and/or control signals over a unidirectional channel (2, 2a) and said data signals over said bidirectional channel from said neighbor unit to said master unit where said signals occur with the same frequency but with a shift in phase and compensating said phase shift by a phase alignment means (3) through which said unidirectional channel (2,2a) is connected directly and said bidirectional channel (20) via a switch (23) to said master unit, said switch controlling the flow of data-in and data-out via said bidirectional channel and where said phase alignment means adjusts the data and/or control signals received from the neighbor unit to exactly the phase of the sending master unit which signals forming the aligned data and/or control signals (AD, AC) for further usage.
generating one or more clock pulses on one single master unit (e.g. A), passing the clock pulses (internal clock A) -together with the control pulses over a unidirectional synchronous channel (1,1a) and with data pulses over a bidirectional channel (20), said pulses having the same frequency and phase as said clock pulses from said master unit to a neighbor unit (e.g. B) where the transferred clock pulses define the local clock(s) (e.g. internal clock B), passing back the clocking and/or control signals over a unidirectional channel (2, 2a) and said data signals over said bidirectional channel from said neighbor unit to said master unit where said signals occur with the same frequency but with a shift in phase and compensating said phase shift by a phase alignment means (3) through which said unidirectional channel (2,2a) is connected directly and said bidirectional channel (20) via a switch (23) to said master unit, said switch controlling the flow of data-in and data-out via said bidirectional channel and where said phase alignment means adjusts the data and/or control signals received from the neighbor unit to exactly the phase of the sending master unit which signals forming the aligned data and/or control signals (AD, AC) for further usage.
3. Method of claim 1 or 2, characterized in that for transferring clocking, data and/or control signals between various units (A,B,C,...; Fig.5) the same master clock (e.g. internal clock A) is used in all the units, the returning signals from the units being applied to phase alignment means (3/1, 3/2, 3/3) in any of the receiving units and provided separately for any of the units connected thereto.
4. Method of claim 1 or 2, characterized in that for transferring clocking, data and/or control signals between a master unit (A) and various neighbor units (B,C,D,E,...; Fig.6) the same master clock (e.g. internal clock A) is used in all the units, the returning signals form said units being applied to separate phase alignment means of said master unit.
5. Method of one or more of the preceding claims 1 to 4, characterized in that for aligning the phase of the returning data and/or control signals the following steps have to be performed:
applying said received data and/or control signals to a buffer array (6; Fig.3B), inputting said signals into successive storage cells (I - N) of the buffer under control of the returning clock pulse signals, reading of saved said data and/or control signals from said buffer storage cells under control of the original master clock signals (e.g. internal clock A)
applying said received data and/or control signals to a buffer array (6; Fig.3B), inputting said signals into successive storage cells (I - N) of the buffer under control of the returning clock pulse signals, reading of saved said data and/or control signals from said buffer storage cells under control of the original master clock signals (e.g. internal clock A)
6. Method of claim 5, characterized by the following further steps:
inputting is controlled by a write pointer (W) which is incremented whenever a clock pulse is received from the returning unit (e.g. B), data and control signals are merged together and stored in said buffer cells indicated by said write pointer, said write pointer is passed as a write copy (WC) to a read control logic (16) of said buffer via a metastability prevention logic (7) for comparison with a read pointer (R) which indicates the buffer cells from which the aligned data and/or control signals (AD,AC) are read and when both said write pointer and said read pointer reach the end of said buffer, they wrap around to the first buffer cells (I).
inputting is controlled by a write pointer (W) which is incremented whenever a clock pulse is received from the returning unit (e.g. B), data and control signals are merged together and stored in said buffer cells indicated by said write pointer, said write pointer is passed as a write copy (WC) to a read control logic (16) of said buffer via a metastability prevention logic (7) for comparison with a read pointer (R) which indicates the buffer cells from which the aligned data and/or control signals (AD,AC) are read and when both said write pointer and said read pointer reach the end of said buffer, they wrap around to the first buffer cells (I).
7. Method of claim 6, characterized by following further steps:
said metastability prevention logic (7) which is used to obtain a stable write copy (WC) across the clock boundary (e.g. internal clock A / received internal clock B) on every clock cycle introduces a series of delays (A), where Dmax >.DELTA.> 1/3 of a clock cycle, for providing at least four different write copies (F,G,H,J) which will be latched by the pulses of the A unit's clock at one time in respective registers (F,G,H,J) in said metastability logic (7) representing said write pointer at four different times and performing the following function to provide a stable write copy (WC) to said read control logic (16), During initialization read pointer R=F=G=H=J, DO WHILE (R <> F) WC=F
END
IF (J-H=G) THEN WC=H
ELSE WC=F, from now on WC always comes from this copy of W (F)
said metastability prevention logic (7) which is used to obtain a stable write copy (WC) across the clock boundary (e.g. internal clock A / received internal clock B) on every clock cycle introduces a series of delays (A), where Dmax >.DELTA.> 1/3 of a clock cycle, for providing at least four different write copies (F,G,H,J) which will be latched by the pulses of the A unit's clock at one time in respective registers (F,G,H,J) in said metastability logic (7) representing said write pointer at four different times and performing the following function to provide a stable write copy (WC) to said read control logic (16), During initialization read pointer R=F=G=H=J, DO WHILE (R <> F) WC=F
END
IF (J-H=G) THEN WC=H
ELSE WC=F, from now on WC always comes from this copy of W (F)
8. Method of claim 7, characterized by the following further steps:
the ignore signal (IGN; Fig. 3A,B) which indicates that invalid data are received form a neighbor unit (e.g.
B) is generated by the read control logic (16) by performing the following function:
IF/R=WC) THEN ( ACTIVATE IGN
R STAYS SAME
ELSE ( DEACTIVATE IGN
( INCREMENT R .
the ignore signal (IGN; Fig. 3A,B) which indicates that invalid data are received form a neighbor unit (e.g.
B) is generated by the read control logic (16) by performing the following function:
IF/R=WC) THEN ( ACTIVATE IGN
R STAYS SAME
ELSE ( DEACTIVATE IGN
( INCREMENT R .
9. Arrangement for performing the method of claim 1 for quasi-synchronously transferring clocking, data, and/or control signals between the master unit (A) and one or more neighbor units (B,..), characterized by a master clock generator (4; Fig. 3A) locally and functionally related to said master unit, the pulse signals of which are used to control said master unit`s internal circuitry, one or more first unidirectional channels (1, 1a, lb) connecting said master unit to said one or more neighbor unit(s) and over which said master clock signals are synchronously transmitted together with data and/or control signals to said one or more neighbor units (B) where said master clock signals are used to control the internal circuitry of said neighbor unit(s), one or more second unidirectional channels (2, 2a, 2b) connecting said neighbor unit(s) to said master unit and over which said clock signals are transferred together with data and/or control signals back to said master unit (A), and one or more phase alignment means (3) related to said master unit and to which said second channel(s) is (are) connected respectively.
10. Arrangement for performing the method of claim 2 for quasi-synchronously transferring clocking, data, and/or control signals between the master unit (A) and one or more neighbor units (B,..), characterized by a master clock generator locally and functionally related to said master unit, the pulse signals of which are used to control said master unit`s internal circuitry, one or more first unidirectional channels (1, 1a;
Fig. 7B) connecting said master unit to said one or more neighbor unit(s) and over which said master clock signals are synchronously transmitted together with said control signals to one or more said neighbor unit(s) (B,...) where said master clock signals are used to control the internal circuitry of said neighbor unit(s), one or more second unidirectional channel(s) (2, 2a) connecting said one or more neighbor unit(s) respectively to said master unit for transferring said clock and control signals from said neighbor unit(s) back to said master unit, one or more bidirectional channel(s) (20) connecting said master unit to said one or more neighbor unit(s) for transferring said data signals from said master unit to said one or more neighbor unit(s) and vice versa, one or more phase alignment means (3) related to said master unit to which said one or more second unidirectional channel(s) is (are) connected respectively, one or more switches (23) related to the master unit and each connecting said bidirectional channel(s) to a different one of said phase alignment means, said switches having two positions (S,R), the one (S) connecting said bidirectional channel to the master unit`s data out line for sending data and the other (R) connecting said bidirectional channel the respective phase alignment means for receiving data from the respective neighbor unit.
Fig. 7B) connecting said master unit to said one or more neighbor unit(s) and over which said master clock signals are synchronously transmitted together with said control signals to one or more said neighbor unit(s) (B,...) where said master clock signals are used to control the internal circuitry of said neighbor unit(s), one or more second unidirectional channel(s) (2, 2a) connecting said one or more neighbor unit(s) respectively to said master unit for transferring said clock and control signals from said neighbor unit(s) back to said master unit, one or more bidirectional channel(s) (20) connecting said master unit to said one or more neighbor unit(s) for transferring said data signals from said master unit to said one or more neighbor unit(s) and vice versa, one or more phase alignment means (3) related to said master unit to which said one or more second unidirectional channel(s) is (are) connected respectively, one or more switches (23) related to the master unit and each connecting said bidirectional channel(s) to a different one of said phase alignment means, said switches having two positions (S,R), the one (S) connecting said bidirectional channel to the master unit`s data out line for sending data and the other (R) connecting said bidirectional channel the respective phase alignment means for receiving data from the respective neighbor unit.
11. Arrangement according to one or more of the preceding claims 1 -11, characterized by phase alignment means (3; Fig. 3B) comprising a buffer array (6) having a plurality of registers (I,...,N) for temporarily storing data and/or control signals coming from the neighbor unit (e.g. B) associated with the phase alignment means, which registers are connected via an address selector (5) to the second unidirectional channel (2, 2b) or when using using a bidirectional channel (20) instead of the unidirectional one, are connected via a switch (23) to this bidirectional channel, a buffer array writing circuitry having address advancing means (10, 15) for controlling said address selector, the address advancing means being advanced by the clock pulses of said associated neighbor unit, a buffer array reading circuitry having address advancing means (16, 17, 18) controlled by the master clock pulses of said master unit, and a metastability prevention logic (7) connected between said buffer array writing circuitry and said buffer array reading circuitry for detecting and indicating erroneous data which have to be ignored.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE91112177.0 | 1991-07-20 | ||
| EP91112177A EP0525221B1 (en) | 1991-07-20 | 1991-07-20 | Quasi-synchronous information transfer and phase alignment means for enabling same |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2074008A1 true CA2074008A1 (en) | 1993-01-21 |
Family
ID=8206961
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002074008A Abandoned CA2074008A1 (en) | 1991-07-20 | 1992-07-16 | Quasi-synchronous information transfer and phase alignment means for enabling same |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US5450572A (en) |
| EP (1) | EP0525221B1 (en) |
| JP (1) | JPH0756616B2 (en) |
| CA (1) | CA2074008A1 (en) |
| DE (1) | DE69115898T2 (en) |
Families Citing this family (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5634041A (en) * | 1992-08-12 | 1997-05-27 | Massachusetts Institute Of Technology | Rationally clocked communication interface |
| DE4314058A1 (en) * | 1993-04-29 | 1994-11-03 | Bayerische Motoren Werke Ag | interface |
| JPH08221315A (en) | 1995-02-15 | 1996-08-30 | Hitachi Ltd | Information processor |
| US5822571A (en) * | 1996-06-05 | 1998-10-13 | Compaq Computer Corporation | Synchronizing data between devices |
| US5974103A (en) * | 1996-07-01 | 1999-10-26 | Sun Microsystems, Inc. | Deterministic exchange of data between synchronised systems separated by a distance |
| US5799175A (en) * | 1996-07-01 | 1998-08-25 | Sun Microsystems, Inc. | Synchronization system and method for plesiochronous signaling |
| US6222380B1 (en) * | 1998-06-15 | 2001-04-24 | International Business Machines Corporation | High speed parallel/serial link for data communication |
| US6473871B1 (en) | 1999-08-31 | 2002-10-29 | Sun Microsystems, Inc. | Method and apparatus for HASS testing of busses under programmable control |
| US6535945B1 (en) | 1999-08-31 | 2003-03-18 | Sun Microsystems, Inc. | Method and apparatus for programmable adjustment of computer system bus parameters |
| US6546507B1 (en) | 1999-08-31 | 2003-04-08 | Sun Microsystems, Inc. | Method and apparatus for operational envelope testing of busses to identify halt limits |
| US6499113B1 (en) | 1999-08-31 | 2002-12-24 | Sun Microsystems, Inc. | Method and apparatus for extracting first failure and attendant operating information from computer system devices |
| US6275077B1 (en) | 1999-08-31 | 2001-08-14 | Sun Microsystems, Inc. | Method and apparatus for programmable adjustment of bus driver propagation times |
| US6609221B1 (en) | 1999-08-31 | 2003-08-19 | Sun Microsystems, Inc. | Method and apparatus for inducing bus saturation during operational testing of busses using a pattern generator |
| US6502212B1 (en) | 1999-08-31 | 2002-12-31 | Sun Microsystems, Inc. | Method and apparatus for bus parameter optimization using probes of system configurations |
| US6321282B1 (en) | 1999-10-19 | 2001-11-20 | Rambus Inc. | Apparatus and method for topography dependent signaling |
| US6987823B1 (en) * | 2000-02-07 | 2006-01-17 | Rambus Inc. | System and method for aligning internal transmit and receive clocks |
| US6898726B1 (en) * | 2000-11-15 | 2005-05-24 | Micron Technology, Inc. | Memory system that sets a predetermined phase relationship between read and write clock signals at a bus midpoint for a plurality of spaced device locations |
| JP2002183692A (en) * | 2000-12-14 | 2002-06-28 | Sony Corp | Ic card and ic card system |
| JP3604372B2 (en) | 2002-02-18 | 2004-12-22 | 沖電気工業株式会社 | Synchronous DRAM control circuit |
| US7477713B2 (en) * | 2004-03-02 | 2009-01-13 | International Business Machines Corporation | method for providing automatic adaptation to frequency offsets in high speed serial links |
| JP2006146430A (en) * | 2004-11-17 | 2006-06-08 | Olympus Corp | Data input/output device, data input/output system and data communication method |
| US7656743B2 (en) * | 2005-11-10 | 2010-02-02 | Qualcomm, Incorporated | Clock signal generation techniques for memories that do not generate a strobe |
| EP2026493A1 (en) * | 2007-08-16 | 2009-02-18 | STMicroelectronics S.r.l. | Method and systems for mesochronous communications in multiple clock domains and corresponding computer program product |
| US20180293025A1 (en) * | 2014-11-10 | 2018-10-11 | Sony Corporation | Interface circuit, memory device, information processing system, and interface circuit controlling method |
| US9946672B2 (en) * | 2015-08-28 | 2018-04-17 | Cirrus Logic, Inc. | Transfer for control data over half-duplex link |
Family Cites Families (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3588703A (en) * | 1967-05-01 | 1971-06-28 | Trw Inc | Phase synchronization system |
| US3970997A (en) * | 1974-08-29 | 1976-07-20 | Honeywell Information Systems, Inc. | High speed peripheral system interface |
| JPS5151247A (en) * | 1974-10-31 | 1976-05-06 | Fujitsu Ltd | KUROTSUKUISOCHOSEIHOHO |
| US4119796A (en) * | 1976-11-01 | 1978-10-10 | Versitron, Inc. | Automatic data synchronizer |
| US4208724A (en) * | 1977-10-17 | 1980-06-17 | Sperry Corporation | System and method for clocking data between a remote unit and a local unit |
| US4677614A (en) * | 1983-02-15 | 1987-06-30 | Emc Controls, Inc. | Data communication system and method and communication controller and method therefor, having a data/clock synchronizer and method |
| US4596026A (en) * | 1983-05-09 | 1986-06-17 | Raytheon Company | Asynchronous data clock generator |
| US4637018A (en) * | 1984-08-29 | 1987-01-13 | Burroughs Corporation | Automatic signal delay adjustment method |
| DE3584751D1 (en) * | 1984-09-21 | 1992-01-09 | Amt Holdings | DATA TRANSFER SYSTEM. |
| US4873703A (en) * | 1985-09-27 | 1989-10-10 | Hewlett-Packard Company | Synchronizing system |
| US4782499A (en) * | 1986-09-29 | 1988-11-01 | Rockwell International Corporation | Automatic alignment of a synchronous data system using a local reference clock and external clock with an unknown delay between the two clocks |
| US4881165A (en) * | 1988-04-01 | 1989-11-14 | Digital Equipment Corporation | Method and apparatus for high speed data transmission between two systems operating under the same clock with unknown and non constant skew in the clock between the two systems |
| US5208913A (en) * | 1988-06-22 | 1993-05-04 | Sharp Kabushiki Kaisha | Buffer memory for synchronizing data transmission and reception between two devices having mutually different operating speeds and operating methods therefor |
| US4943984A (en) * | 1988-06-24 | 1990-07-24 | International Business Machines Corporation | Data processing system parallel data bus having a single oscillator clocking apparatus |
| NL9002426A (en) * | 1990-11-08 | 1992-06-01 | Koninkl Philips Electronics Nv | ELASTIC BUFFER MEMORY. |
-
1991
- 1991-07-20 EP EP91112177A patent/EP0525221B1/en not_active Expired - Lifetime
- 1991-07-20 DE DE69115898T patent/DE69115898T2/en not_active Expired - Fee Related
-
1992
- 1992-06-11 JP JP4151694A patent/JPH0756616B2/en not_active Expired - Lifetime
- 1992-07-16 CA CA002074008A patent/CA2074008A1/en not_active Abandoned
- 1992-07-17 US US07/916,319 patent/US5450572A/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| JPH05189081A (en) | 1993-07-30 |
| EP0525221B1 (en) | 1995-12-27 |
| DE69115898D1 (en) | 1996-02-08 |
| JPH0756616B2 (en) | 1995-06-14 |
| DE69115898T2 (en) | 1996-07-11 |
| EP0525221A1 (en) | 1993-02-03 |
| US5450572A (en) | 1995-09-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2074008A1 (en) | Quasi-synchronous information transfer and phase alignment means for enabling same | |
| US5432823A (en) | Method and circuitry for minimizing clock-data skew in a bus system | |
| US7747888B2 (en) | Technique to create link determinism | |
| US5509038A (en) | Multi-path data synchronizer system and method | |
| US5644604A (en) | Digital phase selector system and method | |
| EP2026493A1 (en) | Method and systems for mesochronous communications in multiple clock domains and corresponding computer program product | |
| JPH02247709A (en) | Method for removing skew | |
| US20090323728A1 (en) | Asynchronous data fifo that provides uninterrupted data flow | |
| EP0606600B1 (en) | Improved single and multistage stage FIFO designs for data transfer synchronizers | |
| US7197582B2 (en) | Low latency FIFO circuit for mixed clock systems | |
| US7519759B2 (en) | Pipeline synchronisation device | |
| JPH07112184B2 (en) | Digital data transfer circuit | |
| US6519688B1 (en) | Read data valid loop-back for high speed synchronized DRAM controller | |
| US5247485A (en) | Memory device | |
| US7792030B2 (en) | Method and system for full-duplex mesochronous communications and corresponding computer program product | |
| KR100817270B1 (en) | Interface device and method for synchronizing data | |
| US5355504A (en) | Self-synchronizing data queues | |
| US7376190B2 (en) | Asynchronous data transmitting apparatus | |
| JP2001060977A (en) | Transmission system | |
| JPH04354219A (en) | Data transmission system | |
| JPH02262739A (en) | Method of transmitting information through bidirectional link, and device to implement this method | |
| KR100278982B1 (en) | Data input / output control circuit | |
| US7076680B1 (en) | Method and apparatus for providing skew compensation using a self-timed source-synchronous network | |
| KR19990056209A (en) | Communication system of status information and control signal between multiple units | |
| KR19980016797A (en) | Synchronous circuit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| FZDE | Discontinued |