US20120294085A1 - Multi-partition architecture for memory - Google Patents
Multi-partition architecture for memory Download PDFInfo
- Publication number
- US20120294085A1 US20120294085A1 US13/561,519 US201213561519A US2012294085A1 US 20120294085 A1 US20120294085 A1 US 20120294085A1 US 201213561519 A US201213561519 A US 201213561519A US 2012294085 A1 US2012294085 A1 US 2012294085A1
- Authority
- US
- United States
- Prior art keywords
- algorithm
- memory
- read
- datalines
- partitions
- 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
Images
Classifications
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C5/00—Details of stores covered by group G11C11/00
- G11C5/02—Disposition of storage elements, e.g. in the form of a matrix array
- G11C5/025—Geometric lay-out considerations of storage- and peripheral-blocks in a semiconductor storage device
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C7/00—Arrangements for writing information into, or reading information out from, a digital store
- G11C7/18—Bit line organisation; Bit line lay-out
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C16/00—Erasable programmable read-only memories
- G11C16/02—Erasable programmable read-only memories electrically programmable
- G11C16/06—Auxiliary circuits, e.g. for writing into memory
- G11C16/26—Sensing or reading circuits; Data output circuits
Definitions
- the present invention relates generally to memories and in particular the present invention relates to memory architecture.
- Partition 100 includes data sectors 102 , and row drivers, column pass gates, predecoders, and timers 104 .
- Sense amps are dedicated by partitions, and circuits 106 to program sense amplifiers for read operations and the like are all dedicated by partition.
- the circuits for each partition can program or read blocks in that partition.
- the circuits necessary for read and program operations are repeated for each partition block, such as partition blocks 202 , 204 , and 206 in a memory device such as device 200 shown in FIG. 2 .
- a flash device with 16 partitions has 16 groupings of circuits, one for each partition, with each circuit individually capable of program or read operations as shown in FIG. 1 .
- FIG. 2 shows a memory 200 with 32 partitions. This memory 202 therefore has 32 program rows as well as 32 sense amp rows. Memories such as this can quickly become very large as density increases.
- a multiple partition memory architecture includes a number of partitions of memory cells, a number of groups of at least two partitions, a common program path and sense path circuit for each of the groups, and multiple separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit.
- a method of reducing circuit overhead in a multiple partition memory includes providing a memory divided into a number of partitions, arranging the partitions into groups of at least two partitions, providing a common program path and sense path circuit for each of the groups, providing separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit, and allowing a concurrent operation on any partition not already in operation.
- a processing system in yet another embodiment, includes a processor and a memory coupled to the processor to store data provided by the processor and to provide data to the processor.
- the memory includes an array of memory cells addressable by address circuitry and input output circuitry, and control circuitry to read, write and erase the memory cells.
- the control circuitry includes a multiple partition memory architecture having a number of partitions of memory cells, a number of groups of at least two partitions, a common program path and sense path circuit for each of the groups, and separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit.
- FIG. 1 is a block diagram of a typical prior art partition architecture
- FIG. 2 is a block diagram of a typical prior art memory using the architecture of FIG. 1 ;
- FIG. 3 is a block diagram of a partition placement according to one embodiment of the present invention.
- FIG. 4 is a block diagram of block placement in a memory according to another embodiment of the present invention.
- FIG. 5 is a block diagram of a portion of a memory according to another embodiment of the present invention.
- FIG. 6 is a diagram of pass gate connections to a partition according to another embodiment of the present invention.
- FIG. 7 is a diagram of a column decode structure according to another embodiment of the present invention.
- the embodiments of the present invention reduce the amount of circuit overhead necessary for concurrent operation. This is accomplished in various embodiments by employing common program path and sense (or read) path circuits, which allow concurrent operation of many different partitions.
- each segment 302 , 304 , 306 , and 308 of the memory 300 has four partitions and a single program/sense amp row.
- Segment 302 comprises in this embodiment partitions labeled 0 , 1 , 2 , and 3 .
- Segment 304 comprises in this embodiment partitions labeled 4 , 5 , 6 , and 7 .
- Segment 306 comprises in this embodiment partitions labeled 8 , 9 , 10 , and 11 .
- Segment 308 comprises in this embodiment partitions labeled 12 , 13 , 14 , and 15 .
- Each segment 302 , 304 , 306 , and 308 has a break between its first two and its last two segments.
- Each segment 302 , 304 , 306 , and 308 has in its break, respectively, a program row 310 , 312 , 314 , and 318 , comprising in one embodiment sense amplifiers and program pathways for each of the four partitions in the segment.
- Each segment (or “stack”) of partitions has its own set of sense amplifiers and program pathways.
- four partitions are “stacked,” but it should be understood that the number of partitions “stacked” is largely dependent upon the size of the die. Any number of stacked partitions could be used.
- the number of partitions which can be accessed by a single program pathway/sense amp row such as rows 310 , 312 , 314 , and 316 , is limited only by the X to Y die ratio of the memory, which is defined by the packaging parameters, and by the speed required of the memory device.
- a number of partitions numbering 2 , 4 , 8 , 16 , or 32 is concurrently accessed using the embodiments of the present invention.
- the space savings of these embodiments is large, since only one set of program/sense circuits are needed where four or more were previously required. Since there is a single program/sense amp row and circuitry for multiple partitions, the circuitry can be designed with a number of features or options that previously would have been too complex and would have occupied too much real estate on the die to be feasibly incorporated into a memory. This is due to the physical number of circuits being reduced. The reduction in the number of circuits results in a much smaller impact on overall die size.
- Block placement for blocks within partitions is shown in one embodiment in FIG. 4 .
- Memory 300 is shown in FIG. 4 with the various partitions 0 - 15 broken down into eight main blocks for each partition.
- blocks 24 - 31 comprise partition 3 .
- eight blocks per partition is shown by way of example and not by way of limitation. A different number of blocks, such as 4 , 16 , 32 or the like is within the scope of the present invention, and is determined in part by current and changing industry specifications.
- FIG. 5 shows a single section 500 of a memory such as memory 300 or the like.
- the section 500 is in this embodiment a 64 Megabit section containing two 32 Megabit subsections 501 and 503 .
- Subsection 501 contains four partitions 502 , 504 , 506 , and 508 , and a sense amplifier section 518 positioned between groupings of two partitions as shown.
- Subsection 503 contains four partitions 508 , 510 , 512 , 514 , and 516 as shown, and a sense amplifier section 520 positioned between groupings of two partitions as shown.
- datalines 522 and 524 for read and datalines 526 and 528 for algorithm (or algo) modes are used, as is shown in greater detail in FIG. 5 .
- the datalines 522 , 524 , 526 , and 528 originate from a program/sense amplifier row (not shown) and cross a plurality of partitions.
- Datalines 522 and 526 cross subsection 501 and partitions 502 , 504 , 506 , and 508 ; and datalines 524 and 528 cross subsection 503 and partitions 510 , 512 , 514 , and 516 .
- Each dataline is locally switched onto a partition's block bitline through pass gates of a particular block if the block is selected, as will be described in further detail below.
- FIG. 5 shows within groups of partitions in a “stack” how spanning is accomplished for the handling of a single sense amplifier row and program row ( 518 and 520 ) for multiple partitions. Rows within partitions and blocks are independently driven, so a typical row decode for a memory is used to decode rows. Columns, however, are linked differently because the datalines for read and algorithm modes span multiple partitions. Using a common sense amplifier with multiple partitions uses datalines that span all partitions in the particular memory subsection. Datalines are dedicated to either a read or an algorithm operation, allowing independent operation of read and algorithm operations. Concurrent operations can then be performed in a single subsection provided the operations are performed in different partitions.
- FIG. 6 shows further detail of the connections of separate read datalines 522 and algorithm datalines 524 to blocks 0 - 7 within one partition, for example partition 506 .
- Each block may be selected for read or algorithm modes as is shown in FIG. 6 . If a block is selected for a read mode, it's pass gate 602 connected to a Zrd signal is turned on, that is the Zrd signal goes high. At the same time, since concurrent operations cannot happen in the same partition, the pass gate 604 connected to a Zalg signal is turned off, that is the Zalg signal goes low. In this situation, the dlcom line 606 is connected to the read dataline 522 .
- a block If a block is selected for an algorithm mode, it's pass gate 604 connected to the Zalg signal is turned on, that is the Zalg signal goes high. At the same time, since concurrent operations cannot happen in the same partition, the pass gate 602 connected to the Zrd signal is turned off, that is the Zrd signal goes low. In this situation, the dlcom line 606 is connected to the algorithm dataline 526 .
- the pass gate select signals Zalg and Zrd are block decoded to go high only if the particular block is selected.
- circuits inside each partition keep the Zrd signal from toggling if an algorithm is currently running in the selected block.
- Y and Z pass gates 608 for each block decode the YZ enable of every column of the block (column decoding).
- the independent devices 602 and 604 are gate connected to the Zrd and Zalg signals for read and algorithm operations, respectively. Control circuitry on the gates of switches 602 and 604 have the lockouts.
- Drivers for the Y pass gates 610 are drivers decoded by block select and partition select in typical fashion. That allows independent Z decoding, and reads on independent datalines without disturb issues.
- FIG. 7 shows one embodiment 700 of a sense amplifier connection to allow simultaneous algorithm verifies and datasheet read operations.
- the read sense amplifiers 702 0 . . . 702 7 are dedicated to read datalines, and one algorithm sense amplifier 704 per I/O is used for internal verifies during a program or erase operation. All circuits related to read and algorithm operations are separated and share only a common ground and supply connection. This prevents any events of one operation from affecting another operation.
- FIG. 7 shows one I/O within a “stack” of multiple partitions. This is a local view of the block that is being accessed for a read operation.
- the grouping of 16 bitlines 706 0 . . . 706 7 , Y pass gates 708 0 . . . 708 7 , and Z pass gates 710 0 . . . 710 7 is repeated for every block top to bottom.
- FIG. 7 shows a grouping of every 16 bitlines 706 0 . . . 706 7 connected to a grouping of Y pass gates 708 0 . . . 708 7 , to Z pass gates 710 0 . . . 710 7 , and then to long datalines dl_read 0 . . . dl_read 7 that span multiple partitions into the grouping of read sense amplifiers 702 0 . . . 702 7 . Further from the grouping of Z pass gates 710 0 . . . 710 7 algorithm datalines dl_alg 0 . . . dl_alg 7 are connected to dedicated algorithm sense amplifier 704 .
- a single select algorithm signal in the group of signals sel_alg ⁇ 0 > . . . sel_alg ⁇ 7 >, which are connected to the gates of the switches 712 0 . . . 712 7 respectively, is selected in this embodiment by an 8-1 multiplexor (not shown) that selects any one of the eight signals sel_alg ⁇ 0 > . . . sel_alg ⁇ 7 >.
- sel_alg ⁇ 7 >maps the selected set of bitlines 706 0 . . . 706 7 to the algorithm sense amplifier 704 .
- the grouping of read datalines dl_read 0 . . . dl_read 7 is connected to read sense amplifiers 702 0 . . . 702 7 . It should be understood that in this embodiment, eight read sense amplifiers are shown. The number of read sense amplifiers is dependent in large part upon the specification of the memory. If the specification of the memory is different, a different number of read sense amplifiers may be used, for example in page mode where a different page size is used.
- the various embodiments of the present invention are amenable to use with memory devices having concurrent operations, including flash memories and the like, and to other devices with simultaneous read and program processes.
- a multi-partition architecture for concurrent operation has been described that includes using a common program path and read paths to cover multiple partitions.
- the use of common program path and read paths for multiple partitions reduces the amount of circuitry die size used, and allows for the maintenance of die size.
Abstract
A multiple partition memory and architecture for concurrent operations reduces circuit overhead by providing a common read sense amplifier and program path for multiple partitions. Long separate datalines for read and algorithm operations allow concurrent operation and blockout of multiple operations in a single block of the memory.
Description
- This Application is a Continuation of U.S. application Ser. No. 10/683,075, titled “MULTI-PARTITION ARCHITECTURE FOR MONEY,” filed Oct. 10, 2003, (Pending) which is commonly assigned and incorporated herein by reference.
- The present invention relates generally to memories and in particular the present invention relates to memory architecture.
- In flash memory products, it is common to design in functionality which supports concurrent operation between partitions or banks of flash memory. Each partition is typically a grouping of several blocks. Each block is individually accessible for program, eras, and read operations. Concurrent operation is dual operation in two different partitions simultaneously. For example, one block in one partition is being read which a different block in a different partition is being programmed or erased. Concurrency operations have been handled in the prior art by having an individual circuit for each partition, as is shown in the
partition architecture 100 ofFIG. 1 .Partition 100 includesdata sectors 102, and row drivers, column pass gates, predecoders, andtimers 104. Sense amps are dedicated by partitions, andcircuits 106 to program sense amplifiers for read operations and the like are all dedicated by partition. The circuits for each partition can program or read blocks in that partition. The circuits necessary for read and program operations are repeated for each partition block, such aspartition blocks device 200 shown inFIG. 2 . A flash device with 16 partitions has 16 groupings of circuits, one for each partition, with each circuit individually capable of program or read operations as shown inFIG. 1 . -
FIG. 2 shows amemory 200 with 32 partitions. Thismemory 202 therefore has 32 program rows as well as 32 sense amp rows. Memories such as this can quickly become very large as density increases. - With concurrent operational memories requiring an individual circuit for each partition block, and with the increasing density of memories and therefore the increasing number of partitions and memory blocks of current memories, individual circuitry for each partition block can cause serious issues with respect to die size. Further, as memory densities continue to increase, such an architecture may effectively limit memory density to below that which is available in order to maintain die size.
- For the reasons stated above, and for other reasons stated below which will become apparent to those skilled in the art upon reading and understanding the present specification, there is a need in the art for a memory architecture that reduces real estate overhead in the memory, and improving functionality while maintaining die size.
- The above-mentioned problems with concurrent operation in memories, increasing memory size and other problems are addressed by the present invention and will be understood by reading and studying the following specification.
- In one embodiment, a multiple partition memory architecture includes a number of partitions of memory cells, a number of groups of at least two partitions, a common program path and sense path circuit for each of the groups, and multiple separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit.
- In another embodiment, a method of reducing circuit overhead in a multiple partition memory includes providing a memory divided into a number of partitions, arranging the partitions into groups of at least two partitions, providing a common program path and sense path circuit for each of the groups, providing separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit, and allowing a concurrent operation on any partition not already in operation.
- In yet another embodiment, a processing system includes a processor and a memory coupled to the processor to store data provided by the processor and to provide data to the processor. The memory includes an array of memory cells addressable by address circuitry and input output circuitry, and control circuitry to read, write and erase the memory cells. The control circuitry includes a multiple partition memory architecture having a number of partitions of memory cells, a number of groups of at least two partitions, a common program path and sense path circuit for each of the groups, and separate read and algorithm datalines, each of the read and algorithm datalines spanning one of the groups and its common program path and sense path circuit.
- Other embodiments are described and claimed.
-
FIG. 1 is a block diagram of a typical prior art partition architecture; -
FIG. 2 is a block diagram of a typical prior art memory using the architecture ofFIG. 1 ; -
FIG. 3 is a block diagram of a partition placement according to one embodiment of the present invention; -
FIG. 4 is a block diagram of block placement in a memory according to another embodiment of the present invention; -
FIG. 5 is a block diagram of a portion of a memory according to another embodiment of the present invention; -
FIG. 6 is a diagram of pass gate connections to a partition according to another embodiment of the present invention; and -
FIG. 7 is a diagram of a column decode structure according to another embodiment of the present invention. - In the following detailed description of the invention, reference is made to the accompanying drawings that form a part hereof, and in which is shown, by way of illustration, specific embodiments in which the invention may be practiced. In the drawings, like numerals describe substantially similar components throughout the several views. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. Other embodiments may be utilized and structural, logical, and electrical changes may be made without departing from the scope of the present invention.
- The following detailed description is, therefore, not to be taken in a limiting sense, and the scope of the present invention is defined only by the appended claims, along with the full scope of equivalents to which such claims are entitled.
- The embodiments of the present invention reduce the amount of circuit overhead necessary for concurrent operation. This is accomplished in various embodiments by employing common program path and sense (or read) path circuits, which allow concurrent operation of many different partitions.
- Referring to
FIG. 3 , a partition placement of architectures in amemory 300 is shown. In this example, eachsegment memory 300 has four partitions and a single program/sense amp row.Segment 302 comprises in this embodiment partitions labeled 0, 1, 2, and 3.Segment 304 comprises in this embodiment partitions labeled 4, 5, 6, and 7.Segment 306 comprises in this embodiment partitions labeled 8, 9, 10, and 11.Segment 308 comprises in this embodiment partitions labeled 12, 13, 14, and 15. Eachsegment segment program row - Each segment (or “stack”) of partitions has its own set of sense amplifiers and program pathways. In the embodiment shown, four partitions are “stacked,” but it should be understood that the number of partitions “stacked” is largely dependent upon the size of the die. Any number of stacked partitions could be used. The number of partitions which can be accessed by a single program pathway/sense amp row such as
rows - The space savings of these embodiments is large, since only one set of program/sense circuits are needed where four or more were previously required. Since there is a single program/sense amp row and circuitry for multiple partitions, the circuitry can be designed with a number of features or options that previously would have been too complex and would have occupied too much real estate on the die to be feasibly incorporated into a memory. This is due to the physical number of circuits being reduced. The reduction in the number of circuits results in a much smaller impact on overall die size.
- Block placement for blocks within partitions is shown in one embodiment in
FIG. 4 .Memory 300 is shown inFIG. 4 with the various partitions 0-15 broken down into eight main blocks for each partition. For example, blocks 24-31comprise partition 3. It should be understood that eight blocks per partition is shown by way of example and not by way of limitation. A different number of blocks, such as 4, 16, 32 or the like is within the scope of the present invention, and is determined in part by current and changing industry specifications. - To accomplish the various embodiments of the present invention, a structure such as that shown in
FIG. 5 is used.FIG. 5 shows asingle section 500 of a memory such asmemory 300 or the like. Thesection 500 is in this embodiment a 64 Megabit section containing two 32Megabit subsections Subsection 501 contains fourpartitions sense amplifier section 518 positioned between groupings of two partitions as shown.Subsection 503 contains fourpartitions sense amplifier section 520 positioned between groupings of two partitions as shown. In order to allow concurrent operations across multiple partitions,separate datalines FIG. 5 . Thedatalines Datalines cross subsection 501 andpartitions cross subsection 503 andpartitions -
FIG. 5 shows within groups of partitions in a “stack” how spanning is accomplished for the handling of a single sense amplifier row and program row (518 and 520) for multiple partitions. Rows within partitions and blocks are independently driven, so a typical row decode for a memory is used to decode rows. Columns, however, are linked differently because the datalines for read and algorithm modes span multiple partitions. Using a common sense amplifier with multiple partitions uses datalines that span all partitions in the particular memory subsection. Datalines are dedicated to either a read or an algorithm operation, allowing independent operation of read and algorithm operations. Concurrent operations can then be performed in a single subsection provided the operations are performed in different partitions. -
FIG. 6 shows further detail of the connections of separate read datalines 522 andalgorithm datalines 524 to blocks 0-7 within one partition, forexample partition 506. Each block may be selected for read or algorithm modes as is shown inFIG. 6 . If a block is selected for a read mode, it'spass gate 602 connected to a Zrd signal is turned on, that is the Zrd signal goes high. At the same time, since concurrent operations cannot happen in the same partition, thepass gate 604 connected to a Zalg signal is turned off, that is the Zalg signal goes low. In this situation, thedlcom line 606 is connected to theread dataline 522. If a block is selected for an algorithm mode, it'spass gate 604 connected to the Zalg signal is turned on, that is the Zalg signal goes high. At the same time, since concurrent operations cannot happen in the same partition, thepass gate 602 connected to the Zrd signal is turned off, that is the Zrd signal goes low. In this situation, thedlcom line 606 is connected to thealgorithm dataline 526. - The pass gate select signals Zalg and Zrd are block decoded to go high only if the particular block is selected. In one embodiment, circuits inside each partition keep the Zrd signal from toggling if an algorithm is currently running in the selected block. There is a lockout based on the algorithm mode in one embodiment, so if an algorithm is running, the block is prevented from any random access. Y and
Z pass gates 608 for each block decode the YZ enable of every column of the block (column decoding). Theindependent devices switches Y pass gates 610 are drivers decoded by block select and partition select in typical fashion. That allows independent Z decoding, and reads on independent datalines without disturb issues. -
FIG. 7 shows oneembodiment 700 of a sense amplifier connection to allow simultaneous algorithm verifies and datasheet read operations. In one embodiment, the read sense amplifiers 702 0 . . . 702 7 are dedicated to read datalines, and onealgorithm sense amplifier 704 per I/O is used for internal verifies during a program or erase operation. All circuits related to read and algorithm operations are separated and share only a common ground and supply connection. This prevents any events of one operation from affecting another operation. -
FIG. 7 shows one I/O within a “stack” of multiple partitions. This is a local view of the block that is being accessed for a read operation. The grouping of 16 bitlines 706 0 . . . 706 7, Y pass gates 708 0 . . . 708 7, and Z pass gates 710 0 . . . 710 7 is repeated for every block top to bottom. -
FIG. 7 shows a grouping of every 16 bitlines 706 0 . . . 706 7 connected to a grouping of Y pass gates 708 0 . . . 708 7, to Z pass gates 710 0 . . . 710 7, and then to long datalines dl_read0 . . . dl_read7 that span multiple partitions into the grouping of read sense amplifiers 702 0 . . . 702 7. Further from the grouping of Z pass gates 710 0 . . . 710 7 algorithm datalines dl_alg0 . . . dl_alg7 are connected to dedicatedalgorithm sense amplifier 704. There is onealgorithm sense amplifier 704 per I/O. A group of switches 712 0 . . . 712 7 are connected between the Z pass gates 710 0 . . . 710 7 andalgorithm sense amplifier 704. A single select algorithm signal in the group of signals sel_alg<0> . . . sel_alg<7>, which are connected to the gates of the switches 712 0 . . . 712 7 respectively, is selected in this embodiment by an 8-1 multiplexor (not shown) that selects any one of the eight signals sel_alg<0> . . . sel_alg<7>. The selection of one of the eight signals sel_alg<0> . . . sel_alg<7>maps the selected set of bitlines 706 0 . . . 706 7 to thealgorithm sense amplifier 704. The grouping of read datalines dl_read0 . . . dl_read7 is connected to read sense amplifiers 702 0 . . . 702 7. It should be understood that in this embodiment, eight read sense amplifiers are shown. The number of read sense amplifiers is dependent in large part upon the specification of the memory. If the specification of the memory is different, a different number of read sense amplifiers may be used, for example in page mode where a different page size is used. - The various embodiments of the present invention are amenable to use with memory devices having concurrent operations, including flash memories and the like, and to other devices with simultaneous read and program processes.
- Conclusion
- A multi-partition architecture for concurrent operation has been described that includes using a common program path and read paths to cover multiple partitions. The use of common program path and read paths for multiple partitions reduces the amount of circuitry die size used, and allows for the maintenance of die size.
- Although specific embodiments have been illustrated and described herein, it will be appreciated by those of ordinary skill in the art that any arrangement, which is calculated to achieve the same purpose, may be substituted for the specific embodiment shown. This application is intended to cover any adaptations or variations of the present invention. Therefore, it is manifestly intended that this invention be limited only by the claims and the equivalents thereof.
Claims (18)
1. A multiple partition memory architecture, comprising:
a plurality of partitions, each partition comprising a plurality of memory blocks of memory cells;
a plurality of groups of at least two partitions;
a plurality of program path and sense path circuits, one common program path and send path circuit for each of the plurality of groups; and
separate pluralities of read and algorithm datalines, each of the plurality of read and algorithm datalines spanning all of one of the plurality of groups.
2. The memory architecture of claim 1 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block.
3. The memory architecture of claim 1 , and further comprising:
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
4. The memory architecture of claim 1 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block; and
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
5. The memory architecture of claim 1 , and further comprising:
a single algorithm sense amplifier for each I/O of the memory architecture, wherein the single algorithm sense amplifier is selectively connectable to one of the algorithm datalines.
6. A flash memory architecture, comprising:
a plurality of partitions, each partition comprising a plurality of memory blocks of memory cells;
a plurality of groups of at least two partitions;
a plurality of program path and sense path circuits, one common program path and sense path circuit for each of the plurality of groups; and
separate pluralities of read and algorithm datalines, each of the plurality of read and algorithm datalines spanning all of one of the plurality of groups.
7. The memory architecture of claim 6 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block.
8. The memory architecture of claim 6 , and further comprising:
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
9. The memory architecture of claim 6 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block; and
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
10. The memory architecture of claim 6 , and further comprising:
a single algorithm sense amplifier for each I/O of the memory architecture, wherein the single algorithm sense amplifier is selectively connectable to one of the algorithm datalines.
11. The memory architecture of claim 6 , wherein the non-volatile memory cells are floating gate memory cells.
12. A flash memory, comprising:
an array of non-volatile memory cells having a multiple partition memory architecture, comprising:
a plurality of partitions, each partition comprising a plurality of memory blocks of memory cells;
a plurality of groups of at least two partitions; and
control circuitry to read, write and erase the memory cells, wherein the control circuitry further comprises:
a plurality of program path and sense path circuits, one common program path and sense path circuit for each of the plurality of groups; and
separate pluralities of read and algorithm datalines, each of the plurality of read and algorithm datalines spanning all of one of the plurality of groups.
13. The memory of claim 12 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block.
14. The memory of claim 12 , and further comprising:
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
15. The memory of claim 12 , and further comprising:
a plurality of read pass gates, each read pass gate selectively connectable between one of the plurality of read datalines and a bitline of a memory block; and
a plurality of algorithm pass gates, each algorithm pass gate selectively connectable between one of the plurality of algorithm datalines and a bitline of a memory block.
16. The memory of claim 12 , and further comprising:
a single algorithm sense amplifier for each I/O of the memory architecture, wherein the single algorithm sense amplifier is selectively connectable to one of the algorithm datalines.
17. A processing system, comprising:
a processor; and
a memory coupled to the processor to store data provided by the processor and to provide data to the processor, the memory comprising:
an array of memory cells addressable by address circuitry and input output circuitry, having a multiple partition memory architecture, comprising:
a plurality of partitions of memory cells;
a plurality of groups of at least two partitions; and
control circuitry to read, write and erase the memory cells;
wherein the control circuitry further comprises:
a plurality of program path and sense path circuits, one common program path and sense path circuit for each of the plurality of groups; and
separate pluralities of read and algorithm datalines, each of the plurality of read and algorithm datalines spanning all of one of the plurality of groups.
18. A memory device comprising:
an array of memory cells having a multiple partition memory architecture, comprising:
a plurality of partitions of memory cells;
a plurality of groups of at least two partitions; and
control circuitry to read, write and erase the memory cells;
wherein the control circuitry further comprises:
a plurality of program path and sense path circuits, one common program path and sense path circuit for each of the plurality of groups; and
separate pluralities of read and algorithm datalines, each of the plurality of read and algorithm datalines spanning all of one of the plurality of groups.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/561,519 US20120294085A1 (en) | 2003-10-10 | 2012-07-30 | Multi-partition architecture for memory |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/683,075 US8233322B2 (en) | 2003-10-10 | 2003-10-10 | Multi-partition memory with separated read and algorithm datalines |
US13/561,519 US20120294085A1 (en) | 2003-10-10 | 2012-07-30 | Multi-partition architecture for memory |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/683,075 Continuation US8233322B2 (en) | 2003-10-10 | 2003-10-10 | Multi-partition memory with separated read and algorithm datalines |
Publications (1)
Publication Number | Publication Date |
---|---|
US20120294085A1 true US20120294085A1 (en) | 2012-11-22 |
Family
ID=34422654
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/683,075 Active 2025-04-21 US8233322B2 (en) | 2003-10-10 | 2003-10-10 | Multi-partition memory with separated read and algorithm datalines |
US13/561,519 Abandoned US20120294085A1 (en) | 2003-10-10 | 2012-07-30 | Multi-partition architecture for memory |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/683,075 Active 2025-04-21 US8233322B2 (en) | 2003-10-10 | 2003-10-10 | Multi-partition memory with separated read and algorithm datalines |
Country Status (1)
Country | Link |
---|---|
US (2) | US8233322B2 (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7876638B2 (en) * | 2007-09-11 | 2011-01-25 | Micron Technology, Inc. | Storing operational information in an array of memory cells |
WO2012168954A1 (en) | 2011-06-10 | 2012-12-13 | Micron Technology Inc. | Apparatus and methods to perform read-while write (rww) operations |
EP3295310A4 (en) | 2015-05-14 | 2018-12-26 | Adesto Technologies Corporation | Concurrent read and reconfigured write operations in a memory device |
US10719237B2 (en) * | 2016-01-11 | 2020-07-21 | Micron Technology, Inc. | Apparatuses and methods for concurrently accessing multiple partitions of a non-volatile memory |
CN112947996B (en) * | 2021-05-14 | 2021-08-27 | 南京芯驰半导体科技有限公司 | Off-chip nonvolatile memory dynamic loading system and method based on virtual mapping |
Family Cites Families (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0593152B1 (en) * | 1992-10-14 | 2000-12-27 | Sun Microsystems, Inc. | Random access memory design |
US5675529A (en) * | 1995-07-07 | 1997-10-07 | Sun Microsystems, Inc. | Fast access memory array |
US5847994A (en) * | 1997-09-08 | 1998-12-08 | Mitsubishi Denki Kabushiki Kaisha | Non-volatile semiconductor memory device having a back ground operation mode |
JP3611280B2 (en) * | 1997-09-30 | 2005-01-19 | シャープ株式会社 | Nonvolatile semiconductor memory device |
US6058461A (en) * | 1997-12-02 | 2000-05-02 | Advanced Micro Devices, Inc. | Computer system including priorities for memory operations and allowing a higher priority memory operation to interrupt a lower priority memory operation |
US5894437A (en) * | 1998-01-23 | 1999-04-13 | Hyundai Elecronics America, Inc. | Concurrent read/write architecture for a flash memory |
US6480929B1 (en) * | 1998-10-31 | 2002-11-12 | Advanced Micro Devices Inc. | Pseudo-concurrency between a volatile memory and a non-volatile memory on a same data bus |
US6377502B1 (en) * | 1999-05-10 | 2002-04-23 | Kabushiki Kaisha Toshiba | Semiconductor device that enables simultaneous read and write/erase operation |
JP3640175B2 (en) * | 2001-04-13 | 2005-04-20 | セイコーエプソン株式会社 | Nonvolatile semiconductor memory device |
US6587917B2 (en) * | 2001-05-29 | 2003-07-01 | Agilent Technologies, Inc. | Memory architecture for supporting concurrent access of different types |
US6614685B2 (en) * | 2001-08-09 | 2003-09-02 | Multi Level Memory Technology | Flash memory array partitioning architectures |
US6529409B1 (en) * | 2001-09-10 | 2003-03-04 | Silicon Storage Technology, Inc. | Integrated circuit for concurrent flash memory with uneven array architecture |
US6643212B1 (en) * | 2002-04-18 | 2003-11-04 | United Memories, Inc. | Simultaneous function dynamic random access memory device technique |
-
2003
- 2003-10-10 US US10/683,075 patent/US8233322B2/en active Active
-
2012
- 2012-07-30 US US13/561,519 patent/US20120294085A1/en not_active Abandoned
Also Published As
Publication number | Publication date |
---|---|
US8233322B2 (en) | 2012-07-31 |
US20050081013A1 (en) | 2005-04-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP1397808B1 (en) | Steering gate and bit line segmentation in non-volatile memories | |
US20040090825A1 (en) | Read-while-write flash memory devices having local row decoder circuits activated by separate read and write signals | |
KR0179361B1 (en) | Bitline segmentation in logic memory arrays c 16/06 | |
US7295485B2 (en) | Memory architecture with advanced main-bitline partitioning circuitry for enhanced erase/program/verify operations | |
US20120294085A1 (en) | Multi-partition architecture for memory | |
US6480417B2 (en) | Global/local memory decode with independent program and read paths and shared local decode | |
US8467252B2 (en) | Word line activation in memory devices | |
US6026021A (en) | Semiconductor memory array partitioned into memory blocks and sub-blocks and method of addressing | |
US8971129B2 (en) | NROM device with reduced power unit | |
US6788600B2 (en) | Non-volatile semiconductor memory | |
US20100070676A1 (en) | Memory Data Bus Placement and Control | |
US6529409B1 (en) | Integrated circuit for concurrent flash memory with uneven array architecture | |
US6631089B1 (en) | Bit line decoding scheme and circuit for dual bit memory array | |
US5982680A (en) | Semiconductor memory | |
US6628536B2 (en) | Semiconductor memory device | |
US7646644B2 (en) | Nonvolatile memory device with multiple references and corresponding control method | |
JPH06195964A (en) | Semiconductor memory | |
KR20020017307A (en) | Semiconductor memory device of multi bank structure | |
KR100487005B1 (en) | High Speed Video Frame Buffer | |
JP2000090676A (en) | Nonvolatile semiconductor memory | |
JP2004348857A (en) | Semiconductor device | |
KR20070061479A (en) | Semiconductor storage device and redundancy method for semiconductor storage device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STCB | Information on status: application discontinuation |
Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION |