Patent Application
20220108742
Certain aspects of the present disclosure generally relate to electronic components and, more particularly, to a compute-in-memory (CIM) cell.
Static random-access memory (SRAM) is a type of memory that uses a flip-flop (FF) to store the value of each bit in a memory cell. In particular, an SRAM is a volatile memory in that the data stored in the memory is lost when the memory is not provided a supply voltage. In some implementations, an SRAM may use compute-in-memory (CIM) circuitry, allowing a logic operation to be performed on data stored in memory simultaneously during a read phase for the memory.
The systems, methods, and devices of the disclosure each have several aspects, no single one of which is solely responsible for its desirable attributes. Without limiting the scope of this disclosure as expressed by the claims which follow, some features will now be discussed briefly. After considering this discussion, and particularly after reading the section entitled “Detailed Description” one will understand how the features of this disclosure provide a compute-in-memory (CIM) cell offering larger analog-to-digital conversion dynamic range and/or improved resolution, as well as increased energy efficiency.
Certain aspects of the present disclosure provide a circuit for in-memory computation. The circuit generally includes a memory cell having a bit-line and a complementary bit-line, a first capacitive element coupled to the bit-line, a second capacitive element coupled to the complementary bit-line, a processing circuit, a first switch coupled between a first input of the processing circuit and the bit-line, and a second switch coupled between a second input of the processing circuit and the complementary bit-line.
Certain aspects of the present disclosure provide a method for in-memory computation. The method generally includes storing charge on at least one of a first capacitive element or a second capacitive element based on a data value stored in a memory cell. The first capacitive element is coupled to a bit-line of the memory cell, and the second capacitive element is coupled to a complementary bit-line of the memory cell. The method also includes sensing, via a processing circuit, the charge stored on the at least one of the first capacitive element or the second capacitive element. A first switch is coupled between a first input of the processing circuit and the bit-line, and a second switch is coupled between a second input of the processing circuit and the complementary bit-line.
To the accomplishment of the foregoing and related ends, the one or more aspects comprise the features hereinafter fully described and particularly pointed out in the claims. The following description and the appended drawings set forth in detail certain illustrative features of the one or more aspects. These features are indicative, however, of but a few of the various ways in which the principles of various aspects may be employed.
So that the manner in which the above-recited features of the present disclosure can be understood in detail, a more particular description, briefly summarized above, may be by reference to aspects, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only certain aspects of this disclosure and are therefore not to be considered limiting of its scope, for the description may admit to other equally effective aspects.
To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. It is contemplated that elements disclosed in one aspect may be beneficially utilized on other aspects without specific recitation.
Certain aspects of the present disclosure relate to a compute-in-memory (CIM) cell implemented using charge sharing (CS) via capacitive elements. For example, the CIM cell may be implemented with a processing system that detects a differential voltage between a read bitline (RBL) and a complementary RBL (RBLB), providing a higher dynamic range with improved resolution for the processing system, as compared to conventional implementations where a single-ended voltage-sensing technique is used.
As illustrated, the FF 209 is coupled between a voltage rail (Vdd) 226 and a reference potential node 228 (e.g., electric ground or Vss). The FF 209 includes a PMOS transistor 208 having a drain coupled to a drain of an NMOS transistor 214, forming part of node N1. The FF 209 also includes a PMOS transistor 210 having a drain coupled to a drain of an NMOS transistor 218, forming part of node N2.
The gates of the PMOS transistor 208 and the NMOS transistor 214 are coupled to the node N2, and the gates of the PMOS transistor 210 and the NMOS transistor 218 are coupled to the node N1, as illustrated. In some implementations, a weight parameter for a neural network or other information may be stored in the FF 209 at nodes N1, N2 of each of the memory cells of the SRAM. The nodes N1, N2 represent the output and complementary output nodes of the FF 209, respectively.
As illustrated, the CIM circuit 200 may also include capacitive elements 270, 272 (respectively labeled “C” and “CB” in
During a computation phase, charge may be stored on the capacitive element 270 or the capacitive element 272 based on a weight parameter stored in the memory cell and activation signal of PCWL and PCWLB. For example, during computing operations, the PG switch 205 and the PG switch 207 may be turned on or off based on activation signal (e.g., via control activation signals PCWL 202 and/or PCWLB 204 output from controller 297), transferring the charge stored at node N1 or N2 to capacitive element 270 or 272. The PG switch 205 and PG switch 207 may then be turned off by controller 297 via the PCWL 202 and/or PCWLB 204. The controller 297 may then set the RWL to logic high to turn on NMOS transistors 216, 220. The NMOS transistors 216, 220 are turned on such that the processing and sensing circuit 234 senses the charge stored on capacitive element 270 and capacitive element 272. In other words, the processing and sensing circuit 234 senses the differential voltage between the RBL 260 and RBLB 262. In certain aspects, after the computation phase, capacitive elements 270 and/or 272 may be discharged via NMOS transistor 216 and/or NMOS transistor 220, respectively, to facilitate further CIM operations. For example, NMOS transistor 216 and NMOS transistor 220 may be turned on while the RBL and RBLB are coupled to a reference potential node by controller 297, effectively discharging the capacitive elements 270, 272.
During a write operation, the logic state at the RBL 260 and RBLB 262 may be set based on a weight parameter to be written to the memory cell. The controller 297 may turn on transistors 216, 220, and the switches 205, 207 may be closed (e.g., turned on) to set the voltages at nodes N1 and N2 of FF 209 based on the weight parameter, in effect writing the weight parameter in the memory cell. The transistors 216, 220 may then be turned off, and the switches 205, 207 may be opened (e.g., turned off).
During a computation phase, switches 205, 207 may be turned on via controller 297 by setting (e.g., via an XOR or XNOR computation activation data) the PCWL and/or PCWLB to logic high and/or logic low (activation), respectively. In some cases, the controller 297 may also couple the RBL and RBLB to a reference potential node and may control transistors 216, 220 to be off. In this configuration, the capacitive element 270 or 272 may be used for computation (e.g., charged) in accordance with the weight parameter stored at nodes N1 and N2 by activation of PCWL and PCWLB. At this stage, controller 297 may open (e.g., turn off) switches 205, 207 and may decouple the RBL and RBLB from the reference potential node. The processing and sensing circuit 234 may then sense the differential voltage between the RBL 260 and RBLB 262 and generate a digital output accordingly. In some aspects, the controller 297 may discharge the capacitive elements 270, 272 by turning on transistors 216, 220 via the RWL to facilitate further computations.
During a write phase, the logic states of the RBL and RBLB may be set based on the weight parameter to be written at nodes N1 and N2. The transistors 216, 220 may be then turned on by the controller 297 setting the RWL to logic high. The switches 205, 207 may be closed (e.g., turned on) to write the weight parameter at nodes N1 and N2. The transistors 216, 220 may be then turned off, and the switches 205, 207 may be opened (e.g., turned off).
The RBL and RBLB may be coupled to multiple memory cells for CIM, the computation outputs of the memory cells being summed and converted to a digital signal by an ADC (e.g., processing and sensing circuit 234). For a single-ended CIM implementation, each CIM memory cell may be implemented using a single capacitive element. The voltage across the capacitive element may be sensed with reference to electric ground (or some other reference potential). As described herein, for a differential ADC implementation, two capacitive elements are used such that a differential voltage between the RBL and RBLB is sensed. The differential voltage may correspond to the difference between the voltages at the RBL and RBLB (e.g., RBL−RBLB). Therefore, the dynamic range of the CIM circuitry implemented with a differential ADC may be doubled, and the overall resolution may be improved, as compared to CIM circuitry implemented with a single-ended ADC.
The operations 400 begin, at block 405, by storing charge on at least one of a first capacitive element (e.g., capacitive element 270) or a second capacitive element (e.g., capacitive element 272) based on a data value (e.g., a weight parameter or other information stored on nodes N1 and N2) stored in a memory cell, where the first capacitive element is coupled to a bit-line (e.g., BL 277) of the memory cell, and where the second capacitive element is coupled to a complementary bit-line (e.g., BLB 279) of the memory cell.
At block 410, the memory system senses, via a processing circuit (e.g., processing and sensing circuit 234), the computed charge stored on the at least one of the first capacitive element or the second capacitive element. In certain aspects, a first switch (e.g., NMOS transistor 216) may be coupled between a first input of the processing circuit and the bit-line, and a second switch (e.g., NMOS transistor 220) may be coupled between a second input of the processing circuit and the complementary bit-line.
In certain aspects, the memory system may also discharge the at least one of the first capacitive element or the second capacitive element via the first switch or the second switch. In some aspects, the memory system may provide the data value to at least one of a read bit-line (e.g., RBL 260) coupled to the first switch or a complementary read bit-line (e.g., RBLB 262) coupled to the second switch, and write the data value in the memory cell by closing at least one of the first switch or the second switch.
In some aspects, the memory cell may include an SRAM memory cell having a FF (e.g., FF 209), a first PG switch (e.g., PG switch 205) coupled between the bit-line and the FF, and a second PG switch (e.g., PG switch 207) coupled between the complementary bit-line and the FF. In this case, the writing of the data value may also include closing the first PG switch and the second PG switch. In some aspects, the sensing of the charge at block 410 may include closing the first switch and the second switch.
Within the present disclosure, the word “exemplary” is used to mean “serving as an example, instance, or illustration.” Any implementation or aspect described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other aspects of the disclosure. Likewise, the term “aspects” does not require that all aspects of the disclosure include the discussed feature, advantage, or mode of operation. The term “coupled” is used herein to refer to the direct or indirect coupling between two objects. For example, if object A physically touches object B and object B touches object C, then objects A and C may still be considered coupled to one another—even if objects A and C do not directly physically touch each other. For instance, a first object may be coupled to a second object even though the first object is never directly physically in contact with the second object. The terms “circuit” and “circuitry” are used broadly and intended to include both hardware implementations of electrical devices and conductors that, when connected and configured, enable the performance of the functions described in the present disclosure, without limitation as to the type of electronic circuits.
The apparatus and methods described in the detailed description are illustrated in the accompanying drawings by various blocks, modules, components, circuits, steps, processes, algorithms, etc. (collectively referred to as “elements”). These elements may be implemented using hardware, for example.
One or more of the components, steps, features, and/or functions illustrated herein may be rearranged and/or combined into a single component, step, feature, or function or embodied in several components, steps, or functions. Additional elements, components, steps, and/or functions may also be added without departing from features disclosed herein. The apparatus, devices, and/or components illustrated herein may be configured to perform one or more of the methods, features, or steps described herein.
It is to be understood that the specific order or hierarchy of steps in the methods disclosed is an illustration of exemplary processes. Based upon design preferences, it is understood that the specific order or hierarchy of steps in the methods may be rearranged. The accompanying method claims present elements of the various steps in a sample order, and are not meant to be limited to the specific order or hierarchy presented unless specifically recited therein.
The previous description is provided to enable any person skilled in the art to practice the various aspects described herein. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects. Thus, the claims are not intended to be limited to the aspects shown herein, but are to be accorded the full scope consistent with the language of the claims, wherein reference to an element in the singular is not intended to mean “one and only one” unless specifically so stated, but rather “one or more.” Unless specifically stated otherwise, the term “some” refers to one or more. A phrase referring to “at least one of” a list of items refers to any combination of those items, including single members. As an example, “at least one of: a, b, or c” is intended to cover at least: a, b, c, a-b, a-c, b-c, and a-b-c, as well as any combination with multiples of the same element (e.g., a-a, a-a-a, a-a-b, a-a-c, a-b-b, a-c-c, b-b, b-b-b, b-b-c, c-c, and c-c-c or any other ordering of a, b, and c). All structural and functional equivalents to the elements of the various aspects described throughout this disclosure that are known or later come to be known to those of ordinary skill in the art are expressly incorporated herein by reference and are intended to be encompassed by the claims. Moreover, nothing disclosed herein is intended to be dedicated to the public regardless of whether such disclosure is explicitly recited in the claims. No claim element is to be construed under the provisions of 35 U.S.C. § 112(f) unless the element is expressly recited using the phrase “means for” or, in the case of a method claim, the element is recited using the phrase “step for.”
It is to be understood that the claims are not limited to the precise configuration and components illustrated above. Various modifications, changes, and variations may be made in the arrangement, operation, and details of the methods and apparatus described above without departing from the scope of the claims.
