Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
1999-12-13
2002-09-24
Bragdon, Reginald G. (Department: 2186)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C711S169000, C713S310000, C713S323000
Reexamination Certificate
active
06457095
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to dynamic random access memory, and more specifically, to low power states of dynamic random access memory.
BACKGROUND
Dynamic random access memory (DRAM) is a general-purpose high-performance memory device suitable for use in a broad range of applications. DRAM allows high bandwidth for multiple, simultaneous, randomly addressed memory transactions.
A particular DRAM may include a pair of low power consumption states for lowering overall system power consumption during periods when data of the DRAM is not accessed. These low power consumption states are the nap and powerdown states. The powerdown (PDN) state is the lowest power state available. In this state the information in the DRAM core is maintained with self-refresh, using an internal timer to refresh the memory. The PDN state has a relatively long exit latency period because of clock resynchronization. An internal clock in the memory is turned off during the PDN state and needs to be resynchronized to an external clock in order to permit normal memory access.
The nap (NAP) state is another low power state in which either self-refresh or refresh-activate (REFA) refresh methods are used to maintain the information in the DRAM core. The NAP state has a shorter exit latency period because the internal clocks in the memory system remain synchronized relative to the external clock signal.
Although the NAP and PDN states are distinct states and have certain differences, in the present application they will often be discussed together as a NAP/PDN state.
FIG. 1
illustrates a state diagram of a prior art memory. State
110
is the NAP/PDN state. The states used to enter into the NAP/PDN state are not illustrated in this figure. The memory may remain in the NAP/PDN state
110
for a period of time. A signal
120
sent by the CPU is received by the memory controller to initiate exit from the NAP/PDN state, moving the memory to the wait for NAP exit delay state
130
. The memory is awakened, clocks are resynchronized, and other “clean-up” steps are taken at this point. The time used for these steps is the “NAP exit delay”, t
NXB
, or “PDN exit delay”, t
pXB
.
After the NAP exit delay or PDN exit delay, the system receives simultaneous quiet times on the row-access-control and column-access-control signal pins of the memory. This moves the memory to the looking-for-packet-frame state
140
.
FIG. 2
illustrates a timing diagram of the prior art system. The clock-to-master (CTM) and clock-from-master (CFM) signals
270
are used by the memory to time data to and from the memory controller. The row-access-control signals
210
and column-access-control signals
220
carry data that identifies the memory location for memory access. The DQAO . . .
8
and DQBO . . .
8
signals
230
are read/write data signals on a data transfer bus.
The SCK signal
240
is a clock signal that is used to time the exit from the NAP/PDN mode. The CMD signal
250
is a command signal used to initiate exiting from the NAP/PDN state. The CMD signal
250
is sampled on both the rising edge and the falling edge of SCK signal
240
. To signal the exit from the NAP/PDN mode, the CMD signal
250
transitions from a 0 on a first falling clock edge
242
to a 1 on the next rising clock edge
244
. Therefore, if after a falling and rising edge of SCK signal
240
there is a “01” on the CMD input, NAP/PDN state will be exited. On a falling edge
242
of the SCK signal
240
, the SIOin signal
260
indicates whether the exit is from a NAP state or a PDN state.
In PDN mode, the CTM/CFM clocks
270
are stopped and must be restarted and stabilized for time t
CE
before a PDN exit command can be sent. In NAP mode, the CTM/CFM clocks
270
are running, and the nap exit command can be sent whenever needed. In both cases, dynamic locked loops (DLLs) in the DRAMs must be restarted and the internal timing circuits of the memory must be resynchronized. After the CTM/CFM clocks
270
become stable, a 0 or 1 is sent on the SIO input
260
on the next falling edge
242
of the SCK signal
240
, for NAP or PDN exit, respectively.
On the next rising edge
244
of the SCK signal
240
, a data signal, PDEV signal
280
, is sent on the DQx pins. The PDEV signal
280
identifies which among several DRAM devices is being woken up from the NAP/PDN state.
Depending upon the DQ select data bit setting for the DRAM device selected to exit NAP/PDN state, the exit delay time begins at either a first falling edge
246
or a second falling edge
248
of SCK signal
240
. At time t
NXB
or t
PXB
—referring either to NAP exit delay or PDN exit delay—after falling edge
246
or
248
, the row-access-control signals
210
and column-access-control signals
220
must enter a quiet state. The quiet cycles
290
on the row-access-control
210
and column-access-control signals
220
must occur exactly t
NXB
or t
PXB
after the appropriate falling edge of the SCK signal
240
. During the quiet cycle, which lasts at least eight clock cycles of the CTM/CFM signals
270
(at least two clock cycles of the SCK signal
240
) no commands may be placed on the row-access-control signal pins
210
or the column-access-control signal pins
220
.
Timing a quiet cycle requires complex processing. If commands appear on the row-access-control signals
210
or column-access-control signals
220
during the required quiet time, the memory may be corrupted. Therefore, a worst case scenario must be taken into consideration when designing the memory controller. In the prior art, the memory itself is not aware of the quiet time scheduling and expects a quiet time
290
at an exact time after the t
NXB
or t
PXB
.
DRAMs are often used in highly pipelined systems. Pipelined systems generally send interrelated and interwoven commands to memory. In order to process a quiet signal
290
at the appropriate time, the commands that would normally be sent during that period must be rescheduled or held for later processing (stalled). All of the commands that are related to the rescheduled commands must be considered. For example, a row-access-control signal
210
may be sent on the row pins. A column-access-control signal
220
must be sent a fixed period after the row signal. This may disrupt pipelining and result in incomplete commands that may result in corrupted data.
One prior art solution is to insert a buffer time prior to the expiration of the NAP/PDN delay. For a time t
buff
prior to the expiration of the delay t
NXB
or t
PXB
no new instructions are sent on the pipeline. The time t
buff
is set such that, prior to the expiration of the delay t
NXB
or t
PXB
, all instructions and data that follow the last pipelined instruction can be completed. Thus, for example, t
buff
is sufficiently long to permit a response for a read query from memory. However, t
buff
inserts a delay into the pipeline and slows down instruction processing.
In the prior art, the quiet cycle is timed simultaneously on the row-access-control signal
210
and column-access-control signal
220
pins. Because the DRAM may address the row-access-control signal
210
and column-access-control signal
220
pins separately, both must be made inactive separately. This requires additional processing in the memory controller. Additionally, because of the cushioning of related commands around the quiet time
290
a longer delay in the signals being sent to the memory may be introduced.
Therefore, a better method of exiting a memory from a low power state would be advantageous.
SUMMARY OF THE INVENTION
A method and apparatus for exiting a memory from a low power state is disclosed. The method includes initiating an exit from the low power state. The method also includes waiting during an exit delay time period. The method further includes scheduling a quiet time command in an addressing pipeline, where the memory transitions from the low power state to a normal power state in response to a quiet time in response to the quiet time command.
REFERENCES:
patent: 5319601 (1994-06-01), Kawata et al.
patent: 5
Blakely , Sokoloff, Taylor & Zafman LLP
Bragdon Reginald G.
Intel Corporation
LandOfFree
Method and apparatus for synchronizing dynamic random access... 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 synchronizing dynamic random access..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for synchronizing dynamic random access... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2896179