Electrical computers and digital data processing systems: input/ – Input/output data processing – Input/output data buffering
Reexamination Certificate
1998-09-11
2001-05-01
Lee, Thomas (Department: 2782)
Electrical computers and digital data processing systems: input/
Input/output data processing
Input/output data buffering
C365S078000, C365S239000
Reexamination Certificate
active
06226698
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to the field of First in First Out (FIFO) buffers, and particularly to a method and apparatus for simultaneously synchronizing information across a boundary between two asynchronous clock domains and for providing an indication of the degree of FIFO fullness to one clock domain in every clock cycle and providing a flag in every clock cycle for the other clock domain indicating the readiness of the FIFO to accept or provide data.
2. Description of the Related Art
FIG. 1
illustrates a prior art approach to transferring data from one clock domain to another clock domain and back. block_A
3
represents circuitry in clock domain A (e.g., signals in block_A
3
are synchronous to clock_A
5
and have reliable data on the rising edge of clock_A
5
). Similarly, block_B
7
represents circuitry and signals that are synchronous to clock_B
9
. A FIFO
11
is provided between clock domains A and B. In this case, data is being transferred from block_A
3
to block_B
7
through the FIFO
11
. Logic
13
, associated with the FIFO, indicates to logic block_A
3
that the FIFO
11
is full. This associated FIFO logic
13
also indicates to block_B
7
that the FIFO
11
is empty. If the FIFO
11
is full, circuitry in block_A
3
stops the transfer of data into the FIFO
11
from block_A
3
to prevent overwrite of valid data. Upon a Full indication
15
, circuitry in block_B
7
begins to transfer data into block_B
7
until an Empty signal
17
is asserted and received by block_B
7
. Block B
7
stops reading data from the FIFO
11
upon an Empty signal
17
so that it will not read invalid data (e.g., stale data).
It is evident in this prior art system that logic in clock domain A will not begin to write to the FIFO until it receives an Empty signal
17
. Similarly, logic in clock domain B will not read from the FIFO
11
until a Full indicator
15
is received. Thus, there is a latency when the logic in clock domain A waits for a completely empty FIFO
11
before writing and another latency when logic in clock domain B waits until the FIFO
11
is completely full before reading the contents of the FIFO
11
.
This prior art approach is satisfactory when both clock domains are relatively slow or when both are equally fast. This prior art is satisfactory when it is not important to have simultaneous FIFO read and write. In other words, for low through put systems, this approach is sufficient. However, in a Sbus ATM (Asynchronous Transfer Mode) system where the ATM is clocked at a much faster rate than the Sbus, prior art methods are no longer sufficient. In order to squeeze Sbus performance to satisfy the demands of AIM, every clock cycle needs to be salvaged. Moreover, when logic in clock domain B (the Sbus side) requires a fast turnaround (e.g., a ATM protocol that requires a 155 megabits per second (Mbps) duplex in an average loaded Sbus 32-bit 32-byte system), this prior art approach becomes inadequate. In a high speed network system, it is imperative to maximize throughput and not wait until the FIFO is 100% full or empty to begin transfer of data into or out of the FIFO. Thus, there is a need for provide varying degrees of fullness or emptiness indications, so as to reduce latency inherent in waiting for a full or empty flag, especially if the FIFO is large.
Furthermore, since logic in clock domain A and logic in clock domain B are asynchronous to each other (e.g., reads and writes to the FIFO are asynchronous), there is a need to dynamically determine at any clock cycle the “true” level of content in the FIFO. For example, if four data words are written to a FIFO by block A and four data words are during the same clock read by block B, there is no net change in the degree of emptiness or fullness of the FIFO.
It is evident that to accurately determine the “true” level of content of the FIFO in any clock cycle, FIFO pointer information from domain B must be synchronized to the clock of domain A, and control information from domain B back to domain A. This information is used to determine the present state or degree of fullness of the FIFO. (e.g., synchronize the signals from the slower clock domain to the faster clock domain).
Information such as control signals, address information, and size information of how much is being written or read from the FIFO may be required to calculate the readiness of the FIFO at each clock cycle (e.g., have enough room or data) to receive or supply from or to an IO DMA slave (e.g., host memory) over the I/O bus.
The prior methods for synchronizing signals from one clock domain to another and back are inadequate for the above-described situations which requires immediate readiness to use the FIFO for data transfer over the IO bus and to maximize throughput for support of high speed networks.
One bit synchronization is not feasible for two reasons. First, as described above, multiple bits of information must be synchronized from one clock domain to another in order to correctly calculate the current level and readiness for data transfer of the FIFO. Second, even if only one bit of information is needed to be synchronized from clock domain A to clock domain B and vice versa, prior art one-bit synchronization schemes have an unacceptable transit time or latency. For example, prior art one-bit synchronization schemes operate on a master and slave principle, and full handshake is required (e.g., a bit of information being synchronized from clock domain A to clock domain B and back from B to A is dependent on information in clock domain B and vice versa.) Thus, a state in clock domain A must not change while the synchronization is taking place (e.g., hold the state of clock domain A while synchronizing a signal from clock domain A to clock domain B).
Furthermore, the now synchronized signal in clock domain B is used for calculations in clock domain B. However, after the calculations and updates have made to the signal, it must now be synchronized back to clock domain A for calculations in clock domain A (e.g., the transit time from one clock domain to another and back exceeds the time constraints of the current application).
Multiple bit synchronization across clock domains introduces unique metastability issues. When data is invalid or uncertain at a particular clock edge (e.g. metastability) in a one-bit synchronization scheme, the true data will eventually reach the correct value (0 or 1) (e.g., a delay is inserted and the correct value is presented at the next clock cycle). Since clock A and clock B are completely asynchronous to each other (e.g., the clocks may be different in phase and frequency), data at a particular clock edge of the synchronizing clock may be unpredictable and unreliable in some instances.
However, when multiple bits are simultaneously being synchronized across clock domains (e.g., a plurality of bits in an address or size information), this metastability problem is exacerbated since two or more bits having unreliable data results in a completely wrong address or size and not simply a delay. In other words, the multi-bit value cannot be trusted at the rising clock edge (i.e., metastability). Thus, prior art methods that require multiple bit synchronization across clock domains typically use techniques (e.g., Grey Code or Johnson Counter) to ensure that a multiple-bit value can only change one bit per clock cycle. These schemes are expensive to implement (e.g., it wastes bits to implement the Grey Code especially if the multiple-bit value is greater than four bits), and the resulting hardware, especially if there are many signals being synchronized across the clock boundaries, is difficult to test and debug.
Thus, there is also a need for a method and apparatus to provide synchronization logic that minimizes the number of bits being synchronized between clock domains and yet provides a true indicator of readiness for data transfer of the FIFO at each clock cycle.
SUMMARY OF THE INVENTION
An interface circuit that is coupled between a first circuit
Cen Ling
Yeung Louise Y.
Blakely & Sokoloff, Taylor & Zafman
Lee Thomas
Peyton Tammara
Sun Microsystems Inc.
LandOfFree
Method and apparatus for dynamically calculating degrees of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for dynamically calculating degrees of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for dynamically calculating degrees of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2514499