`VARIOUS METHODS OF DRAM REFRESH
`
`TECHNICAL
`NOTE
`
`VARIOUS METHODS
`OF DRAM REFRESH
`
`This article was originally published in 1994.
`
`INTRODUCTION
`DRAM refresh is the topic most misunderstood by
`designers due to the many ways refresh can be accom-
`plished. This article addresses the most often asked
`questions about refresh. The two basic means of per-
`forming refresh, distributed and burst, are explained
`first, followed by the various ways to accomplish re-
`fresh: RAS#-ONLY REFRESH, CAS#-BEFORE-RAS# RE-
`FRESH and HIDDEN REFRESH.
`
`STANDARD AND EXTENDED REFRESH
`DRAMs are often referred to as either “standard
`refresh” or “extended refresh.” Dividing the specified
`refresh time by the number of cycles required will
`determine if the DRAM is a standard refresh or an
`extended refresh device. If the result is 15.6µs, it is a
`standard refresh device, while a result of 125µs indi-
`cates an extended refresh device.
`Table 1 lists some of the standard DRAMs and their
`refresh specifications.
`
`Table 1
`Standard DRAMs and
`Refresh Specifications
`REFRESH NUMBER OF REFRESH
`TIME
`CYCLES
`RATE
`16ms
`1,024
`15.6µs
`8ms
`512
`15.6µs
`64ms
`512
`125µs
`
`DRAM
`4 Meg x 1
`256K x 16
`256K x 16
`(L version)
`4 Meg x 4
`(2K)
`4 Meg x 4
`(4K)
`
`on before repeating the task. When not being re-
`freshed, the DRAM can be read from or written to.
`
`BURST REFRESH
`Refresh may be achieved in a burst method by
`performing a series of refresh cycles, one right after the
`other until all rows have been accessed. During refresh
`other commands are not allowed. Below is a drawing
`representing burst and distributed refresh.
`For example: a 4 Meg x 1 requires 1,024 consecutive
`refresh cycles, each of which will use 130ns (tRC) for a
`70ns device:
`1,024 cycles ¥ 130ns = 133,120ns = 0.133ms
`16ms - 0.133ms = 15.867ms
`Approximately 0.13ms would be spent performing
`refresh, and the remaining 15.87ms could be spent
`reading and writing; then burst refresh would occur
`again, and so on.
`Distributed refresh is the more common of the two
`refresh categories. The DRAM controller is set up to
`perform a refresh cycle every 15.6µs. Usually, this
`means the controller allows the current cycle to be
`completed and then holds off all instructions while a
`refresh is performed on the DRAM. The requested cycle
`is then allowed to resume.
`
`REFRESH CYCLES
`There are different cycles you can use to refresh
`DRAMs, all of which can be used in a distributed or
`burst method. There are three types listed in a standard
`data sheet:
`• RAS#-ONLY REFRESH
`• CAS#-BEFORE-RAS# REFRESH
`• HIDDEN REFRESH
`
`32ms
`
`64ms
`
`2,046
`
`4,096
`
`15.6µs
`
`15.6µs
`
`Distributed
`Refresh
`
`Burst
`Refresh
`
`DISTRIBUTED REFRESH
`Distributing the refresh cycles so that they are evenly
`spaced is known as distributed refresh. To perform
`distributed refresh on a standard DRAM, execute a
`refresh cycle every 15.6µs such that all rows are turned
`
`Time
`
`Each pulse represents
`a refresh cycle
`
`Required time to
`complete refresh of all rows
`
`Figure 1
`Burst and Destributed Refresh
`
`TN-04-30
`DT30.p65 – Rev. 2/99
`
`1
`
`Micron Technology, Inc., reserves the right to change products or specifications without notice.
`©1999, Micron Technology, Inc.
`
`Netlist Ex 2023
`Samsung v Netlist
`IPR2022-00996
`
`
`
`TN-04-30
`VARIOUS METHODS OF DRAM REFRESH
`
`RAS#-ONLY REFRESH
`To perform a RAS#-ONLY REFRESH, a row address is
`put on the address lines and then RAS# is dropped.
`When RAS# falls, that row will be refreshed and as long
`as CAS# is held HIGH, the DQs will remain open. (See
`Figure 2.)
`It is the DRAM controller’s function to provide the
`addresses to be refreshed and make sure that all rows are
`being refreshed in the appropriate amount of time. The
`row order of refreshing does not matter; what is impor-
`tant is that each row be refreshed in the specified
`amount of time.
`
`CAS#-BEFORE-RAS# REFRESH
`CAS#-BEFORE-RAS# REFRESH, also known as CBR
`REFRESH, is a frequently used method of refresh be-
`cause it is easy to use and offers the advantage of a
`power savings. A CBR REFRESH cycle is performed by
`dropping CAS# and then dropping RAS#. One refresh
`cycle will be performed each time RAS# falls. WE# must
`be held HIGH while RAS# falls. The DQs will remain
`open during the cycle.
`Here’s how CBR REFRESH works. The die contains
`an internal counter which is initialized to a random
`count when the device is powered up. Each time a CBR
`REFRESH is performed, the device refreshes a row based
`on the counter, and then the counter is incremented.
`When CBR REFRESH is performed again, the next row
`is refreshed and the counter is incremented. The counter
`will automatically wrap and continue when it reaches
`the end of its count. There is no way to reset the
`
` One refresh cycle when RAS# falls
`
`counter. The user does not have to supply or keep track
`of row addresses. A drawing of one CBR REFRESH cycle
`is shown in Figure 3. CAS# must be held LOW before
`and after RAS# falls to meet tCSR and tCHR. Figure 4
`shows three CBR REFRESH cycles. In this drawing, CAS#
`stays LOW and only RAS# toggles. Every time RAS# falls
`a refresh cycle is performed. CAS# may be toggled each
`time, but it’s not necessary.
`
`CBR POWER SAVINGS
`Since CBR REFRESH uses the internal counter and
`not an external address, the address buffers are pow-
`ered-down. For power-sensitive applications, this can
`be a benefit because there is no additional current used
`in switching address lines on a bus, nor will the DRAMs
`pull extra power if the address voltage is at an interme-
`diate state.
`
`CBR REFRESH IS EASY TO USE
`Since CBR REFRESH uses its own internal counter,
`there is not a concern about the controller having to
`supply the refresh addresses. Virtually all DRAMs sup-
`port CBR REFRESH and the 15.6µs refresh rate, so you
`can design for CBR REFRESH at the distributed rate of
`15.6µs and plug in many different DRAMs without
`having to worry about refresh. For example, the 4 Meg
`x 4 comes in two versions:
`• 2,048 cycles in 32ms
`• 4,096 cycles in 64ms
`
`tRC
`
`tRAS
`
`tRP
`
`tCRP
`
`tRPC
`
`tASR
`
`tRAH
`
`ROW
`
`OPEN
`
`Figure 2
`RAS#-Only Refresh
`
`DON’T CARE
`
`UNDEFINED
`
`IH
`IL
`
`VV
`
`IH
`IL
`
`VV
`
`IH
`IL
`
`VV
`
`RAS#
`
`CASL#
`and CASH#
`
`ADDR
`
`OH
`OL
`
`VV
`
`Q
`
`TN-04-30
`DT30.p65 – Rev. 2/99
`
`2
`
`Micron Technology, Inc., reserves the right to change products or specifications without notice.
`©1999, Micron Technology, Inc.
`
`Netlist Ex 2023
`Samsung v Netlist
`IPR2022-00996
`
`
`
`TN-04-30
`VARIOUS METHODS OF DRAM REFRESH
`
`If CBR REFRESH is used, simply maintain the stan-
`dard 15.6µs refresh rate. If RAS#-ONLY REFRESH is
`used, addresses must be supplied as follows:
`• A0-A10 for the 2,048 cycle refresh
`• A0-A11 for the 4,096 cycle refresh
`
`HIDDEN REFRESH
`In HIDDEN REFRESH, the user does a READ or
`WRITE cycle and then, leaving CAS# LOW, brings RAS#
`HIGH (for minimum of tRP) and then LOW. Since CAS#
`
`was LOW before RAS# went LOW, the part will execute
`a CBR REFRESH. In a READ cycle the output data will
`remain valid during the CBR REFRESH. The refresh is
`not “hidden” in the sense that you can hide the time it
`takes to refresh; instead, it is hidden in the sense that
`data-out will stay on the lines while performing the
`function. READ and HIDDEN REFRESH cycles will take
`the same amount of time: tRC. The two cycles together
`take 2 x tRC. If we were to do a READ and then follow
`it with a standard CBR REFRESH (instead of a HIDDEN
`
` One refresh cycle
`
`tRP
`
`tRAS
`
`tRPC
`
`tCPN
`
`tCSR
`
`tCHR
`
`tRPC
`
`tWRP
`
`tWRH
`
`OPEN
`
`IH
`IL
`
`VV
`
`RAS#
`
`IH
`IL
`
`VV
`
`CAS#
`
`DQ
`
`IH
`IL
`
`VV
`
`WE#
`
`Figure 3
`One CAS#-Before-RAS# Refresh Cycle
`
` One refresh cycle
`
` One refresh cycle
`
` One refresh cycle
`
`tRP
`
`tRAS
`
`tRP
`
`tRAS
`
`tRP
`
`tRAS
`
`tRPC
`
`tCPN
`
`tCSR
`
`tWRP
`
`tWRH
`
`tWRP
`
`tWRH
`
`tWRP
`
`tWRH
`
`OPEN
`
`IH
`IL
`
`VV
`
`RAS#
`
`IH
`IL
`
`VV
`
`CAS#
`
`DQ
`
`IH
`IL
`
`VV
`
`WE#
`
`DON’T CARE
`
`UNDEFINED
`
`Figure 4
`Three CAS#-Before-RAS# Refresh Cycles
`
`TN-04-30
`DT30.p65 – Rev. 2/99
`
`3
`
`Micron Technology, Inc., reserves the right to change products or specifications without notice.
`©1999, Micron Technology, Inc.
`
`Netlist Ex 2023
`Samsung v Netlist
`IPR2022-00996
`
`
`
`TN-04-30
`VARIOUS METHODS OF DRAM REFRESH
`
`REFRESH), this would take the same amount of time: 2
`x tRC.
`Figure 5 shows a READ followed by a HIDDEN
`REFRESH. Figure 6 shows a READ followed by a stan-
`dard CBR REFRESH. The only difference between the
`two is that data-out is valid during the HIDDEN RE-
`FRESH.
`
`SUMMARY
`Three different cycles exist to perform refresh on a
`standard DRAM: RAS#-ONLY REFRESH, CBR REFRESH,
`and HIDDEN REFRESH. Each cycle can be used in a
`burst or distributed method, whichever best fits the
`designer’s needs. It is strongly urged that CBR REFRESH
`be used to refresh the DRAM. Future DRAMs will most
`likely require CBR REFRESH only.
`
`(READ)
`
`tRAS
`
`tRP
`
`(REFRESH)
`tRAS
`
`tRP
`
`tOFF
`
`tCRP
`
`tRCD
`
`tRSH
`
`tASR
`
`tRAH
`
`tASC
`
`tCAH
`
`IH
`IL
`
`VV
`
`IH
`IL
`
`VV
`
`RAS#
`
`CAS#
`
`COLUMN
`tAA
`tRAC
`tCAC
`tCLZ
`
`ROW
`
`IH
`IL
`
`VV
`
`ADDR
`
`OPEN
`
`VALID DATA
`
`OPEN
`
`tOE
`
`tORD
`
`tOD
`
`Figure 5
`READ Cycle Followed by Hidden Refresh
`
`(READ)
`
`tRAS
`
`tRP
`
`(REFRESH)
`tRAS
`
`tRP
`
`tCRP
`
`tRCD
`
`tRSH
`
`tCSR
`
`tCHR
`
`tASR
`
`tRAH
`
`tASC
`
`tCAH
`
`ROW
`
`COLUMN
`tAA
`tRAC
`tCAC
`tCLZ
`
`tOFF
`
`OPEN
`
`OPEN
`
`tOE
`
`tORD
`
`VALID DATA
`
`tOD
`
`Figure 6
`READ Cycle Followed by CBR REFRESH
`
`Micron is a registered trademark of Micron Technology, Inc.
`
`DON’T CARE
`
`UNDEFINED
`
`IH
`IL
`
`VV
`
`WE#
`
`IOH
`IOL
`
`VV
`
`DQx
`
`IH
`IL
`
`VV
`
`OE#
`
`IH
`IL
`
`VV
`
`IH
`IL
`
`VV
`
`IH
`IL
`
`VV
`
`RAS#
`
`CAS#
`
`ADDR
`
`IH
`IL
`
`VV
`
`WE#
`
`IOH
`IOL
`
`VV
`
`DQx
`
`IH
`IL
`
`VV
`
`OE#
`
`TN-04-30
`DT30.p65 – Rev. 2/99
`
`4
`
`Micron Technology, Inc., reserves the right to change products or specifications without notice.
`©1999, Micron Technology, Inc.
`
`Netlist Ex 2023
`Samsung v Netlist
`IPR2022-00996
`
`