The present disclosure relates generally to semiconductor memory and methods, and more particularly, to apparatuses and methods related to implementing scatter and gather operations in a memory array.
Memory devices are typically provided as internal, semiconductor, integrated circuits in computers or other electronic systems. There are many different types of memory including volatile and non-volatile memory. Volatile memory can require power to maintain its data (e.g., host data, error data, etc.) and includes random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), synchronous dynamic random access memory (SDRAM), and thyristor random access memory (TRAM), among others. Non-volatile memory can provide persistent data by retaining stored data when not powered and can include NAND flash memory, NOR flash memory, and resistance variable memory such as phase change random access memory (PCRAM), resistive random access memory (RRAM), and magnetoresistive random access memory (MRAM), such as spin torque transfer random access memory (STT RAM), among others.
Electronic systems often include a number of processing resources (e.g., one or more processors), which may retrieve and execute instructions and store the results of the executed instructions to a suitable location. A processor can comprise a number of functional units such as arithmetic logic unit (ALU) circuitry, floating point unit (FPU) circuitry, and/or a combinatorial logic block, for example, which can be used to execute instructions by performing logical operations such as AND, OR, NOT, NAND, NOR, and XOR, and invert (e.g., inversion) logical operations on data (e.g., one or more operands). For example, functional unit circuitry may be used to perform arithmetic operations such as addition, subtraction, multiplication, and/or division on operands via a number of logical operations.
A number of components in an electronic system may be involved in providing instructions to the functional unit circuitry for execution. The instructions may be generated, for instance, by a processing resource such as a controller and/or host processor. Data (e.g., the operands on which the instructions will be executed) may be stored in a memory array that is accessible by the functional unit circuitry. The instructions and/or data may be retrieved from the memory array and sequenced and/or buffered before the functional unit circuitry begins to execute instructions on the data. Furthermore, as different types of operations may be executed in one or multiple clock cycles through the functional unit circuitry, intermediate results of the instructions and/or data may also be sequenced and/or buffered.
In many instances, the processing resources (e.g., processor and/or associated functional unit circuitry may be external to the memory array, and data is accessed via a bus between the processing resources and the memory array to execute a set of instructions. Processing performance may be improved in a processor-in-memory device, in which a processor may be implemented internal and/or near to a memory (e.g., directly on a same chip as the memory array). A processing-in-memory device may save time by reducing and/or eliminating external communications and may also conserve power.
The present disclosure includes apparatuses and methods related to scatter/gather in a memory array. In one embodiment, the apparatus comprises a memory device coupled to a host via a data bus and a control bus. The memory device includes an array of memory cells and sensing circuitry coupled to the array via a plurality of sense lines. The sensing circuitry includes sense amplifiers and a compute component configured to implement logical operations.
A memory controller is coupled to the array and sensing circuitry. The memory controller is configured to receive a block of instructions, including a gather and/or scatter operation from a channel controller associated with the host and memory device. The memory controller is further configured to subdivide the block of instructions so that operations before, between and after a local gather operation can be executed locally on the memory device.
Scatter and/or gather (hereinafter “scatter/gather”), also known as vectored Input/Output (I/O), is a technique used in applications accessing large data sets. When applied to large data sets, memory array access patterns generated by scatter/gather operations can often show very low ordering and locality of successive accesses. Data accesses can be dispersed all over the memory array, which can be inefficient. Embodiments of the present disclosure provide an apparatus and methods for implementing an efficient scatter/gather technique involving a memory device having processing-in-memory (PIM) capabilities.
Scatter can be generally expressed as:
for (i=0;i<N;i++)
y[idx[i]]=x[i]
and gather can be generally expressed as:
for (i=0;i<N;i++)
x[i]=y[idx[i]]
where i is an index (e.g., counter, pointer) and N represents a number of indices used to store a quantity of data in a memory array. The expression idx[i] represents an index vector comprised of indices. As such, the requested data, x[i], is a function (y) of the index vector. In this manner, the index vector contains information by which pieces of the requested data are stored/found in the memory array.
For small-sized index vectors, the operations to extract and process the indices do not pose a significant challenge for modem computing system and memory device. However, when the aggregate size of the index vectors become larger than the caching ability of the computing system, performance of the scatter/gather operations decreases significantly due to data being fetched into cache, used a small number of times (e.g., once) and thereafter flushed out of cache having to be re-read into cache again later. However where large quantities of data are being requested, this inefficiency is more than just an inadequate cache size. Giga Updates Per Second (GUPS) is an example of one recognized benchmark to measure performance of a computing system and memory device for scatter/gather operations.
The index vector (e.g., idx[i]) can be thought of as being “pathological” in that the pattern of data elements comprising the requested data, and to which a pattern of index vector elements corresponds, can impact the efficiency and performance of scatter/gather operations. Previous approaches to scatter/gather operations utilized simple “tricks” to optimize the overall scatter/gather operation. However, as the amount of data being requested increases, the quantity of possible patterns in the index vector can also increase. The index vector can take on many index patterns that can confound some simple tricks previously utilized to optimize scatter/gather operations. In order to appreciate the improved scatter/gather techniques and apparatus for implementing such techniques, a discussion of a memory device having PIM capabilities, and associated host, follows.
According to various embodiments, performance of scatter/gather operations involving a memory device having PIM capabilities can distribute implementation of a scatter/gather operation over multiple sensing circuitries that can implement logical operations within the memory array. Scatter/gather operations involving a memory device having PIM capabilities can be completed in less time and using less power. Some time and power advantage can be realized by reducing the amount of data that is moved around a computing system to process the requested memory array operations (e.g., reads, writes, etc.).
A number of embodiments of the present disclosure can provide improved parallelism and/or reduced power consumption in association with performing compute functions as compared to previous systems such as previous PIM systems and systems having an external processor (e.g., a processing resource located external from a memory array, such as on a separate integrated circuit chip). For instance, a number of embodiments can provide for performing fully complete compute functions such as integer add, subtract, multiply, divide, and CAM (content addressable memory) functions without transferring data out of the memory array and sensing circuitry via a bus (e.g., data bus, address bus, control bus), for instance. Such compute functions can involve performing a number of logical operations (e.g., logical functions such as AND, OR, NOT, NOR, NAND, XOR, etc.). However, embodiments are not limited to these examples. For instance, performing logical operations can include performing a number of non-Boolean logic operations such as copy, compare, destroy, etc.
In previous approaches, data may be transferred from the array and sensing circuitry (e.g., via a bus comprising input/output (I/O) lines) to a processing resource such as a processor, microprocessor, and/or compute engine, which may comprise ALU circuitry and/or other functional unit circuitry configured to perform the appropriate logical operations. However, transferring data from a memory array and sensing circuitry to such processing resource(s) can involve significant power consumption. Even if the processing resource is located on a same chip as the memory array, significant power can be consumed in moving data out of the array to the compute circuitry, which can involve performing a sense line (which may be referred to herein as a digit line or data line) address access (e.g., firing of a column decode signal) in order to transfer data from sense lines onto I/O lines (e.g., local I/O lines), moving the data to the array periphery, and providing the data to the compute function.
Furthermore, the circuitry of the processing resource(s) (e.g., compute engine) may not conform to pitch rules associated with a memory array. For example, the cells of a memory array may have a 4F2 or 6F2 cell size, where “F” is a feature size corresponding to the cells. As such, the devices (e.g., logic gates) associated with ALU circuitry of previous PIM systems may not be capable of being formed on pitch with the memory cells, which can affect chip size and/or memory density, for example. A number of embodiments of the present disclosure include sensing circuitry formed on pitch with an array of memory cells and capable of performing compute functions such as gather and scatter operations local to the array of memory cells.
In the following detailed description of the present disclosure, reference is made to the accompanying drawings that form a part hereof, and in which is shown by way of illustration how one or more embodiments of the disclosure may be practiced. These embodiments are described in sufficient detail to enable those of ordinary skill in the art to practice the embodiments of this disclosure, and it is to be understood that other embodiments may be utilized and that process, electrical, and/or structural changes may be made without departing from the scope of the present disclosure. As used herein, designators such as “N”, “M”, etc., particularly with respect to reference numerals in the drawings, indicate that a number of the particular feature so designated can be included. As used herein, “a number of” a particular thing can refer to one or more of such things (e.g., a number of memory arrays can refer to one or more memory arrays). A “plurality of” is intended to refer to more than one of such things.
The figures herein follow a numbering convention in which the first digit or digits correspond to the drawing figure number and the remaining digits identify an element or component in the drawing. Similar elements or components between different figures may be identified by the use of similar digits. For example, 206 may reference element “06” in
System 100 includes a host 110 coupled (e.g., connected) to memory device 120, which includes a memory array 130. Host 110 can be a host system such as a personal laptop computer, a desktop computer, a digital camera, a smart phone, or a memory card reader, among various other types of hosts. Host 110 can include a system motherboard and/or backplane and can include a number of processing resources (e.g., one or more processors, microprocessors, or some other type of controlling circuitry). The system 100 can include separate integrated circuits or both the host 110 and the memory device 120 can be on the same integrated circuit. The system 100 can be, for instance, a server system and/or a high performance computing (HPC) system and/or a portion thereof. Although the example shown in
For clarity, the system 100 has been simplified to focus on features with particular relevance to the present disclosure. The memory array 130 can be a DRAM array, SRAM array, STT RAM array, PCRAM array, TRAM array, RRAM array, NAND flash array, and/or NOR flash array, for instance. The array 130 can comprise memory cells arranged in rows coupled by access lines (which may be referred to herein as word lines or select lines) and columns coupled by sense lines, which may be referred to herein as data lines or digit lines. Although a single array 130 is shown in
The memory device 120 includes address circuitry 142 to latch address signals provided over a data bus 156 (e.g., an I/O bus) through I/O circuitry 144. Status, exception, and other data information can be provided from the memory controller 140 on the memory device 120 to a channel controller 143 via a high speed interface (HSI) (both shown in
Memory controller 140, e.g., bank control logic and/or sequencer, decodes signals provided by a control bus 154 from the host 110. These signals can include chip enable signals, write enable signals, and address latch signals that are used to control operations performed on the memory array 130, including data read, data write, and data erase operations. In various embodiments, the memory controller 140 is responsible for executing instructions from the host 110. The memory controller 140 can be a state machine, a sequencer, or some other type of controller. The controller 140 can control shifting data (e.g., right or left) in an array, e.g., memory array 130.
Examples of the sensing circuitry 150 are described further below. For instance, in a number of embodiments, the sensing circuitry 150 can comprise a number of sense amplifiers and a number of compute components, which may serve as, and be referred to herein as, an accumulator and can be used to perform logical operations (e.g., on data associated with complementary data lines).
In a number of embodiments, the sensing circuitry 150 can be used to perform logical operations using data stored in array 130 as inputs and store the results of the logical operations back to the array 130 without transferring data via a sense line address access (e.g., without firing a column decode signal). As such, various compute functions can be performed using, and within, sensing circuitry 150 rather than (or in association with) being performed by processing resources external to the sensing circuitry (e.g., by a processor associated with host 110 and/or other processing circuitry, such as ALU circuitry, located on device 120 (e.g., on controller 140 or elsewhere)).
In various previous approaches, data associated with an operand, for instance, would be read from memory via sensing circuitry and provided to external ALU circuitry via I/O lines (e.g., via local I/O lines and/or global I/O lines). The external ALU circuitry could include a number of registers and would perform compute functions using the operands, and the result would be transferred back to the array via the I/O lines. In contrast, in a number of embodiments of the present disclosure, sensing circuitry 150 is configured to perform logical operations on data stored in memory array 130 and store the result back to the memory array 130 without enabling an I/O line (e.g., a local I/O line) coupled to the sensing circuitry 150. The sensing circuitry 150 can be formed on pitch with the memory cells of the array. Additional peripheral sense amplifier and/or logic 170 can be coupled to the sensing circuitry 150.
As such, in a number of embodiments, circuitry external to array 130 and sensing circuitry 150 is not needed to perform compute functions as the sensing circuitry 150 can perform the appropriate logical operations to perform such compute functions without the use of an external processing resource. Therefore, the sensing circuitry 150 may be used to compliment and/or to replace, at least to some extent, such an external processing resource (or at least the bandwidth consumption of such an external processing resource).
However, in a number of embodiments, the sensing circuitry 150 may be used to perform logical operations (e.g., to execute instructions) in addition to logical operations performed by an external processing resource (e.g., host 110). For instance, host 110 and/or sensing circuitry 150 may be limited to performing only certain logical operations and/or a certain number of logical operations.
Enabling an I/O line can include enabling (e.g., turning on) a transistor having a gate coupled to a decode signal (e.g., a column decode signal) and a source/drain coupled to the I/O line. However, embodiments are not limited to not enabling an I/O line. For instance, in a number of embodiments, the sensing circuitry (e.g., 150) can be used to perform logical operations without enabling column decode lines of the array; however, the local I/O line(s) may be enabled in order to transfer a result to a suitable location other than back to the array 130 (e.g., to an external register).
As shown in
For example, each of the plurality of banks, e.g., Bank 0, . . . , Bank 7, in the plurality of memory devices 120-1, . . . , 120-N can include address circuitry 142 to latch address signals provided over a data bus 156 (e.g., an I/O bus) through I/O circuitry 144. Status and/or exception information can be provided from the memory controller 140 on the memory device 120 to the channel controller 143, using the OOB bus 157, which in turn can be provided from the plurality of memory devices 120-1, . . . , 120-N to the host 110. For each of the plurality of banks, e.g., Bank 0, . . . , Bank 7, address signals can be received through address circuitry 142 and decoded by a row decoder 146 and a column decoder 152 to access the memory array 130. Data can be read from memory array 130 by sensing voltage and/or current changes on the data lines using sensing circuitry 150. The sensing circuitry 150 can read and latch a page (e.g., row) of data from the memory array 130. The I/O circuitry 144 can be used for bi-directional data communication with host 110 over the data bus 156. The write circuitry 148 is used to write data to the memory array 130 and the OOB bus 157 can be used to report status and exception information to the channel controller 143.
As described next in connection with
For clarity, the system 100 has been simplified to focus on features with particular relevance to the present disclosure. The memory array 130 can be a DRAM array, SRAM array, STT RAM array, PCRAM array, TRAM array, RRAM array, NAND flash array, and/or NOR flash array, for instance. The array 130 can comprise memory cells arranged in rows coupled by access lines (which may be referred to herein as word lines or select lines) and columns coupled by sense lines, which may be referred to herein as data lines or digit lines. Although a single array 130 is shown in
As in
A host 110 including an integrated (in 111 dashed line form) channel controller 143 (e.g., scaler processor), or alternatively a channel controller 143 integrated in module form, can send a block of instructions that include individual instructions (e.g., commands) for at least one of a gather operation and a scatter operation to a plurality of memory devices 120-1, . . . , 120-N (shown in
In one embodiment, the channel controller 143 can extract an individual instruction comprising indices (e.g., individual vectors) from the index vector 200, and load the indices to a local idx buffer 202. The buffer 161 of the channel controller 143 shown in
Typically, instructions executed in the memory device (e.g., memory chips), or portion of the memory device (e.g., bank), can be limited in scope to data contained in the memory device or portion of the memory device. Since it cannot be certain that a gather operation is confined to data solely within a particular portion of the memory device 120 (e.g., bank), a bank arbiter can orchestrate the gather operation among a plurality of memory controllers 140 associated with a plurality of memory portions (e.g., banks). As shown in
The memory controller 140 to a given bank can de-reference locations of indices (e.g., individual vectors) in the block of instructions. The memory controller 140 is configured such that individual instructions in the received block of instructions can be subdivided and sent to a memory controller 140 such that operations before, between, and after local gather and/or scatter operations can be executed locally in a particular portion of a given memory device 120. The individual instructions subdivided out of the block of instructions can be referred to herein as a local instruction set.
The memory controller 140 can manage execution of the local instruction set on the associated portion of the memory device (e.g., bank). Execution can be performed by the sensing circuitry 150, e.g., PIM RAM execution units 250-1, . . . , 250-N in respective banks, performing logical operations and/or functions as the same is described herein. The memory controller 140 can monitor the local instruction set for completion in order to perform the gather operation at an appropriate time to ensure local, in-order execution. In this way, the memory controller 140 can manage an appropriate result for a particular data request, e.g., x-source vector in DRAM 201. Further, the memory controller can ensure that all other operations of a portion of the memory device are complete before starting a gather operation.
According to various embodiments, the memory controller 140 manages the gather operation to obtain portions of requested data, e.g., x-source vector in DRAM 201, from different portions of the memory device and/or different portions within a bank, e.g., subarray. As the reader will appreciate, the bank arbiter 145 can manage the gather operation from among each memory controller 140 to obtain portions of requested data from among the plurality of banks, e.g., Bank 0, . . . , Bank 7, of the memory device 120. The channel controller 143 can iterate through the indices of the index vector to “stream” 257 a collection of data from the portions of the memory device 120 comprising the requested data, e.g., using the bank arbiter 145, high speed interface 141 and the out-of-band (OOB) bus 157 to field the requested data to the channel controller 143 and/or return the requested data to the host 110. For example, the channel controller 143 can allocate buffers 161, shown as individual bank command buffers (BANK CMD BUFFER) 204-1, 204-2, . . . , 204-N in
As the channel controller 143 encounters a gather operation in a block of memory device instructions requested by the host 110, the channel controller 143 can implement the gather operation by reading a local index (idx) buffer 202 block by block (e.g., index register storing the idx vector 200). Each memory controller 140 associated with banks of the memory device 120 can locally store a moderately-sized buffer of relevant data corresponding to a portion of the index vector 200 distributed by the bank arbiter 145 to that particular bank. For example, each memory controller 140 may include a storage area as a buffer and may include a register. The memory controller 140 obtains the portion of requested data, e.g., the x-source vector portion, pointed to by an individual index, relevant to that bank. The memory controller 140 can group requests for portions of the requested data together by location, e.g., subarray, such that the portion of the memory device, e.g., bank, can execute instructions in an efficient manner to gather the necessary data and return it 257 to the channel controller 143 via the OOB bus 157 (also referred to as channel).
According to various embodiments, a portion of the requested data associated with a particular portion of the memory device 120, e.g., bank, is returned to the channel controller 143 via the out-of-band (OOB) bus 157 to avoid resource contention with further reads of the index vector 200. The OOB bus 157 can also be used to avoid resource contention with further writing of data to the y-target vector 203.
According to some embodiments, as data is returned to the channel controller via the OOB bus 157, each portion of the requested data can be tagged 261 as to its intended destination within the requested data, e.g., within the y-target vector 203. The channel controller 143 can assemble the portions of data obtained from various locations into a coherent writes to burst into target locations, e.g., y-target vector in DRAM 203.
Using the OOB channel to retrieve portions of the requested data avoids use of the data bus 156 shown in
A scatter operation differs from a gather operation in that the portions of the memory device 120, e.g., sensing circuitry 150, operate to store portions of the requested data into the y-target vector 203, e.g., the host requests each memory device 120 to store the requested data into the y-target vector in DRAM. In the scatter operation, the channel controller 143 will be reading locations to be written to from the idx vector 200 and the data to be written from the x-source vector 201. The channel controller 143 will then group write requests by locality, e.g., bin indices according to PIM RAM structures (e.g., individual PIM RAM execution units (memory controllers 140)) and create individual instructions, stored in local command bank buffers 204-1, . . . , 204-N, to send those individual instructions to appropriate memory controllers 140 in respective memory devices 120-1, . . . , 120-N. The channel controller 143 will provide a stream of instructions to be executed by the sensing circuity 150 of the respective portions of the memory device 120-1, . . . , 120-N that contains embedded data and specific locations such that each memory device 120 can execute these local write requests independent of the host 110 and/or channel controller 143, e.g., can be executed “offline”. As used herein, “offline” refers to operations implemented by the sensing circuitries 150 of the respective portions of memory devices 120-1, . . . , 120-N rather than by a data read and/or write involving the data bus 156. With respect to scatter operations, the OOB bus 157 can be used to ensure the local individual instructions to each memory device 120/memory controller 140, e.g., local writes, are completed.
In one example, the HSI 141 is a first in first out (FIFO) style interface to the bank arbiter 145 of a processor in memory (PIM), e.g., memory device 120. The bank arbiter side can be a read FIFO interface with an empty bit, as well as clock and read signals, and a data out bus representing the bank, e.g., 121-1, . . . , 121-M, and row address, as well as the operation type, activate, pre-charge, and/or auto refresh for the PIM 120. Control logic and sequencers as part of memory controller 140 of each bank 121-1, . . . , 121-M (e.g., Bank 0, . . . , Bank 7) can each have a FIFO coming into the bank arbiter 145 for pushing status information. The bank arbiter 145 pulls status information off each of these FIFOs (e.g., 16 FIFOs for 2 each by 8 banks) and pushes the status information back to the host 110 via the HSI 141 and channel controller 143. The FIFOs may be checked in round robin fashion using a counter.
The HSI or status channel interface, e.g., OOB, 141 can be a dedicated interface between the memory device 120 and the channel controller 143. By way of example, and not by way of limitation, this interface may be composed of sixteen (16) data outputs, 4 data strobes, and a rank enable input. In this example, when the memory device 120 is configured as a x64 DQ part, the status output pins are bused to each memory device 120-1, . . . , 120-M on a given channel of the channel controller 143. For example, the channel controller 143 may be equipped with four (4) rank enable outputs per channel. Rank Enable zero (0) may be connected to all memory devices 120-1, . . . , 120-M on the first rank, Rank Enable one (1) can be connected to all memory devices on the second rank, and so on as shown in the various example embodiments in
In this case, the bandwidth of the status channel is divided just as the bandwidth of a double data rate (DDR) interface is divided, being effectively time domain multiplexed to avoid contention on the OOB 441. Each memory device 420, e.g., 420-1, 420-2, . . . , 420-4, is granted access to the uni-directional status bus 441 in turn as its rank enable input is activated. When x64 mode is selected (fuse option or test option), each memory device 420 may transmit one data packet at an active edge of the rank enable input. Status output bits 449 can be streamed out, 1 bit per clock cycle, synchronized to the SQS strobe pins 451. The first bit may be a preamble bit. In one example, if the preamble bit is a logic 0, the memory device, e.g., 420-1, 420-2, . . . , 420-4, will transmit a 128 bit status message on the 16 DQs on eight (8) successive clocks. At the end of this message the SQ and SQS pins will transition to a tri-state mode, and the interface circuitry will wait for the next active edge on the rank enable input to transmit additional messages. If there are no messages waiting in a memory controller's (e.g., memory controller 440 of a given memory device 420) output queue, the sense of the preamble bit will be logic 1. The SQ and SQS drivers will be turned off on the next successive clock. The memory device will go back to waiting for another active edge on the rank enable input.
In this example, the rank enable inputs 547 for the first rank are all tied together. Thus they will all begin transmitting their status messages on the lower eight (8) SQ pins 549. The protocol is largely the same as in the x64 case, with data being sent following a preamble on successive active clock edges of the SQSs. One deviation from the prior example is that since the data bus is half as wide for each memory device 520, the message length will increase from eight (8) data clocks to sixteen (16), to accommodate the same 128 bit message size.
A memory cell comprises a storage element (e.g., capacitor) and an access device (e.g., transistor). For instance, a first memory cell comprises transistor 702-1 and capacitor 703-1, and a second memory cell comprises transistor 702-2 and capacitor 703-2, etc. In this example, the memory array 730 is a DRAM array of 1T1C (one transistor one capacitor) memory cells. In a number of embodiments, the memory cells may be destructive read memory cells (e.g., reading the data stored in the cell destroys the data such that the data originally stored in the cell is refreshed after being read).
The cells of the memory array 730 can be arranged in rows coupled by word lines 704-X (Row X), 704-Y (Row Y), etc., and columns coupled by pairs of complementary sense lines (e.g., data lines DIGIT(n−1)/DIGIT(n−1)_, DIGIT(n)/DIGIT(n)_, DIGIT(n+1)/DIGIT(n+1)_). The individual sense lines corresponding to each pair of complementary sense lines can also be referred to as data lines 705-1 (D) and 705-2 (D_) respectively. Although only one pair of complementary data lines are shown in
Memory cells can be coupled to different data lines and/or word lines. For example, a first source/drain region of a transistor 702-1 can be coupled to data line 705-1 (D), a second source/drain region of transistor 702-1 can be coupled to capacitor 703-1, and a gate of a transistor 702-1 can be coupled to word line 704-X. A first source/drain region of a transistor 702-2 can be coupled to data line 705-2 (D_), a second source/drain region of transistor 702-2 can be coupled to capacitor 703-2, and a gate of a transistor 702-2 can be coupled to word line 704-Y. The cell plate, as shown in
The memory array 730 is coupled to sensing circuitry 750 in accordance with a number of embodiments of the present disclosure. In this example, the sensing circuitry 750 comprises a sense amplifier 706 and a compute component 731 corresponding to respective columns of memory cells (e.g., coupled to respective pairs of complementary data lines). The sense amplifier 706 can be coupled to the pair of complementary sense lines 705-1 and 705-2. The compute component 731 can be coupled to the sense amplifier 706 via pass gates 707-1 and 707-2. The gates of the pass gates 707-1 and 707-2 can be coupled to logical operation selection logic 713.
The logical operation selection logic 713 can be configured to include pass gate logic for controlling pass gates that couple the pair of complementary sense lines un-transposed between the sense amplifier 706 and the compute component 731 (as shown in
The sense amplifier 706 can be operated to determine a data value (e.g., logic state) stored in a selected memory cell. The sense amplifier 706 can comprise a cross coupled latch, which can be referred to herein as a primary latch. In the example illustrated in
In operation, when a memory cell is being sensed (e.g., read), the voltage on one of the data lines 705-1 (D) or 705-2 (D_) will be slightly greater than the voltage on the other one of data lines 705-1 (D) or 705-2 (D_). An ACT signal and the RNL* signal can be driven low to enable (e.g., fire) the sense amplifier 706. The data lines 705-1 (D) or 705-2 (D_) having the lower voltage will turn on one of the PMOS transistor 729-1 or 729-2 to a greater extent than the other of PMOS transistor 729-1 or 729-2, thereby driving high the data line 705-1 (D) or 705-2 (D_) having the higher voltage to a greater extent than the other data line 705-1 (D) or 705-2 (D_) is driven high.
Similarly, the data line 705-1 (D) or 705-2 (D_) having the higher voltage will turn on one of the NMOS transistor 727-1 or 727-2 to a greater extent than the other of the NMOS transistor 727-1 or 727-2, thereby driving low the data line 705-1 (D) or 705-2 (D_) having the lower voltage to a greater extent than the other data line 705-1 (D) or 705-2 (D_) is driven low. As a result, after a short delay, the data line 705-1 (D) or 705-2 (D_) having the slightly greater voltage is driven to the voltage of the supply voltage VCC through source transistor 711, and the other data line 705-1 (D) or 705-2 (D_) is driven to the voltage of the reference voltage (e.g., ground) through the sink transistor 713. Therefore, the cross coupled NMOS transistors 727-1 and 727-2 and PMOS transistors 729-1 and 729-2 serve as a sense amplifier pair, which amplify the differential voltage on the data lines 705-1 (D) and 705-2 (D_) and operate to latch a data value sensed from the selected memory cell. As used herein, the cross coupled latch of sense amplifier 706 may be referred to as a primary latch 715.
Embodiments are not limited to the sense amplifier 706 configuration illustrated in
The sense amplifier 706 can, in conjunction with the compute component 731, be operated to perform various logical operations using data from an array as input. In a number of embodiments, the result of a logical operation can be stored back to the array without transferring the data via a data line address access (e.g., without firing a column decode signal such that data is transferred to circuitry external from the array and sensing circuitry via local I/O lines). As such, a number of embodiments of the present disclosure can enable performing logical operations and compute functions associated therewith using less power than various previous approaches. Additionally, since a number of embodiments eliminate the need to transfer data across I/O lines in order to perform compute functions (e.g., between memory and discrete processor), a number of embodiments can enable an increased parallel processing capability as compared to previous approaches.
The sense amplifier 706 can further include equilibration circuitry 714, which can be configured to equilibrate the data lines 705-1 (D) and 705-2 (D). In this example, the equilibration circuitry 714 comprises a transistor 724 coupled between data lines 705-1 (D) and 705-2 (D). The equilibration circuitry 714 also comprises transistors 725-1 and 725-2 each having a first source/drain region coupled to an equilibration voltage (e.g., VDD/2), where VDD is a supply voltage associated with the array. A second source/drain region of transistor 725-1 can be coupled data line 705-1 (D), and a second source/drain region of transistor 725-2 can be coupled data line 705-2 (D). Gates of transistors 724, 725-1, and 725-2 can be coupled together, and to an equilibration (EQ) control signal line 726. As such, activating EQ enables the transistors 724, 725-1, and 725-2, which effectively shorts data lines 705-1 (D) and 705-2 (D_) together and to the an equilibration voltage (e.g., VCC/2).
Although
As described further below, in a number of embodiments, the sensing circuitry (e.g., sense amplifier 706 and compute component 731) can be operated to perform a selected logical operation and initially store the result in one of the sense amplifier 706 or the compute component 731 without transferring data from the sensing circuitry via an I/O line (e.g., without performing a data line address access via activation of a column decode signal, for instance).
Performance of logical operations (e.g., Boolean logical functions involving data values) is fundamental and commonly used. Boolean logic functions are used in many higher level functions. Consequently, speed and/or power efficiencies that can be realized with improved logical operations, which can translate into speed and/or power efficiencies of higher order functionalities. Described herein are apparatuses and methods for performing logical operations without transferring data via an input/output (I/O) line and/or without transferring data to a control component external to the array. Depending on memory array architecture, the apparatuses and methods for performing the logical operations may not require amplification of a sense line (e.g., data line, digit line, bit line) pair.
As shown in
The gates of the pass gates 807-1 and 807-2 can be controlled by a logical operation selection logic signal, Pass. For example, an output of the logical operation selection logic can be coupled to the gates of the pass gates 807-1 and 807-2. The compute component 831 can comprise a loadable shift register configured to shift data values left and right.
According to the embodiment illustrated in
The sensing circuitry shown in
According to various embodiments, the logical operation selection logic 813 can include four logic selection transistors: logic selection transistor 862 coupled between the gates of the swap transistors 842 and a TF signal control line, logic selection transistor 852 coupled between the gates of the pass gates 807-1 and 807-2 and a TT signal control line, logic selection transistor 854 coupled between the gates of the pass gates 807-1 and 807-2 and a FT signal control line, and logic selection transistor 864 coupled between the gates of the swap transistors 842 and a FF signal control line. Gates of logic selection transistors 862 and 852 are coupled to the true sense line through isolation transistor 850-1 (having a gate coupled to an ISO signal control line). Gates of logic selection transistors 864 and 854 are coupled to the complementary sense line through isolation transistor 850-2 (also having a gate coupled to an ISO signal control line).
Data values present on the pair of complementary sense lines 805-1 and 805-2 can be loaded into the compute component 831 via the pass gates 807-1 and 807-2. The compute component 831 can comprise a loadable shift register. When the pass gates 807-1 and 807-2 are OPEN, data values on the pair of complementary sense lines 805-1 and 805-2 are passed to the compute component 831 and thereby loaded into the loadable shift register. The data values on the pair of complementary sense lines 805-1 and 805-2 can be the data value stored in the sense amplifier 806 when the sense amplifier is fired. The logical operation selection logic signal, Pass, is high to OPEN the pass gates 807-1 and 807-2.
The ISO, TF, TT, FT, and FF control signals can operate to select a logical function to implement based on the data value (“B”) in the sense amplifier 806 and the data value (“A”) in the compute component 831. In particular, the ISO, TF, TT, FT, and FF control signals are configured to select the logical function to implement independent from the data value present on the pair of complementary sense lines 805-1 and 805-2 (although the result of the implemented logical operation can be dependent on the data value present on the pair of complementary sense lines 805-1 and 805-2. That is, the ISO, TF, TT, FT, and FF control signals select the logical operation to implement directly since the data value present on the pair of complementary sense lines 805-1 and 805-2 is not passed through logic to operate the gates of the pass gates 807-1 and 807-2.
Additionally,
The logical operation selection logic signal Pass can be activated (e.g., high) to OPEN the pass gates 807-1 and 807-2 (e.g., conducting) when the ISO control signal line is activated and either the TT control signal is activated (e.g., high) with data value on the true sense line is “1” or the FT control signal is activated (e.g., high) with the data value on the complement sense line is “1.”
The data value on the true sense line being a “1” OPENs logic selection transistors 852 and 862. The data value on the complimentary sense line being a “1” OPENs logic selection transistors 854 and 864. If the ISO control signal or either the respective TT/FT control signal or the data value on the corresponding sense line (e.g., sense line to which the gate of the particular logic selection transistor is coupled) is not high, then the pass gates 807-1 and 807-2 will not be OPENed by a particular logic selection transistor.
The logical operation selection logic signal PassF can be activated (e.g., high) to OPEN the swap transistors 842 (e.g., conducting) when the ISO control signal line is activated and either the TF control signal is activated (e.g., high) with data value on the true sense line is “1,” or the FF control signal is activated (e.g., high) with the data value on the complement sense line is “1.” If either the respective control signal or the data value on the corresponding sense line (e.g., sense line to which the gate of the particular logic selection transistor is coupled) is not high, then the swap transistors 842 will not be OPENed by a particular logic selection transistor.
The Pass* control signal is not necessarily complementary to the Pass control signal. It is possible for the Pass and Pass* control signals to both be activated or both be deactivated at the same time. However, activation of both the Pass and Pass* control signals at the same time shorts the pair of complementary sense lines together, which may be a disruptive configuration to be avoided.
The sensing circuitry illustrated in
Logic Table 9-1 illustrated in
Via selective control of the continuity of the pass gates 807-1 and 807-2 and the swap transistors 842, each of the three columns of the upper portion of Logic Table 9-1 can be combined with each of the three columns of the lower portion of Logic Table 9-1 to provide 3×3=9 different result combinations, corresponding to nine different logical operations, as indicated by the various connecting paths shown at 975. The nine different selectable logical operations that can be implemented by the sensing circuitry 850 are summarized in Logic Table 9-2 illustrated in
The columns of Logic Table 9-2 illustrated in
While example embodiments including various combinations and configurations of sensing circuitry, sense amplifiers, compute component, dynamic latches, isolation devices, and/or shift circuitry have been illustrated and described herein, embodiments of the present disclosure are not limited to those combinations explicitly recited herein. Other combinations and configurations of the sensing circuitry, sense amplifiers, compute component, dynamic latches, isolation devices, and/or shift circuitry disclosed herein are expressly included within the scope of this disclosure.
Although specific embodiments have been illustrated and described herein, those of ordinary skill in the art will appreciate that an arrangement calculated to achieve the same results can be substituted for the specific embodiments shown. This disclosure is intended to cover adaptations or variations of one or more embodiments of the present disclosure. It is to be understood that the above description has been made in an illustrative fashion, and not a restrictive one. Combination of the above embodiments, and other embodiments not specifically described herein will be apparent to those of skill in the art upon reviewing the above description. The scope of the one or more embodiments of the present disclosure includes other applications in which the above structures and methods are used. Therefore, the scope of one or more embodiments of the present disclosure should be determined with reference to the appended claims, along with the full range of equivalents to which such claims are entitled.
In the foregoing Detailed Description, some features are grouped together in a single embodiment for the purpose of streamlining the disclosure. This method of disclosure is not to be interpreted as reflecting an intention that the disclosed embodiments of the present disclosure have to use more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive subject matter lies in less than all features of a single disclosed embodiment. Thus, the following claims are hereby incorporated into the Detailed Description, with each claim standing on its own as a separate embodiment.
This application is a Continuation of U.S. application Ser. No. 15/669,300, filed Aug. 4, 2017, which is a Continuation of International Application Number PCT/US2016/015027, filed Jan. 27, 2016, which claims the benefit to U.S. Provisional Application 62/112,843, filed Feb. 6, 2015, the entire contents of which are incorporated herein by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
62112843 | Feb 2015 | US |
Number | Date | Country | |
---|---|---|---|
Parent | 15669300 | Aug 2017 | US |
Child | 16536941 | US | |
Parent | PCT/US2016/015027 | Jan 2016 | US |
Child | 15669300 | US |