# US 2021/0279039 A1 · Time-shared compute-in-memory bitcell

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0279039 A1
- Document ID: US20210279039A1
- Status: Published application
- Kind: A1
- Application number: 16/811,480
- Filed: 2020-03-06
- Published: 2021-09-09
- Earliest claimed: 2020-03-06
- Assignees: QUALCOMM Incorporated
- Applicants: QUALCOMM Incorporated
- Inventors: Ankit SRIVASTAVA
- Web page: https://patentgrep.com/patent/US20210279039A1

## CPC

- G06F 7/57: Arithmetic logic units [ALU], i.e. arrangements or devices for performing two or more of the operations covered by groups G06F7/483 – G06F7/556 or for performing logical operations {(G06F7/49, G06F7/491 take precedence)}
- G06G 7/163: using a variable impedance controlled by one of the input signals, variable amplification or transfer function {(G06G7/161, G06G7/162 take precedence)}
- G11C 7/1006: {Data managing, e.g. manipulating data before writing or reading out, data bus switches or control circuits therefor}
- G11C 11/412: using field-effect transistors only
- G11C 11/419: Read-write [R-W] circuits
- G11C 11/54: using elements simulating biological cells, e.g. neuron
- H10B 10/12: {comprising a MOSFET load element}
- H10D 89/10: Integrated device layouts
- G06N 20/00: Machine learning
- H03M 1/38: sequentially only, e.g. successive approximation type (converting more than one bit per step H03M1/14)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06G: ANALOGUE COMPUTERS (analogue optical computing devices G06E3/00)
- G11C: STATIC STORES (semiconductor memory devices H10B)
- H10B: ELECTRONIC MEMORY DEVICES
- H10D: INORGANIC ELECTRIC SEMICONDUCTOR DEVICES
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H03M: CODING; DECODING; CODE CONVERSION IN GENERAL (using fluidic means F15C4/00; optical analogue/digital converters G02F7/00; coding, decoding or code conversion, specially adapted for particular applications, see the relevant subclasses, e.g. G01D, G01R, G06F, G06T, G09G, G10L, G11B, G11C, H04B, H04L, H04M, H04N; ciphering or deciphering for cryptography or other purposes involving the need for secrecy G09C)

## Abstract

A compute-in-memory array is provided that includes a set of compute-in-memory bitcells that time share a shared capacitor connected between the set of compute-in-memory bitcells and a read bit line.

## Claims

### Claim 1

A compute-in-memory bitcell array, comprising:
a read bit line;
a shared capacitor having a first plate connected to the read bit line;
a first compute-in-memory bitcell that includes a first pair of cross-coupled inverters having a first output node for a first stored bit and includes a first transmission gate connected between the first output node and a second plate of the shared capacitor, the first transmission gate being configured to close in response to a first input bit being true and to open in response to the first input bit being false; and
a second compute-in-memory bitcell that includes a second pair of cross-coupled inverters having a second output node for a second stored bit and includes a second transmission gate connected between the second output node and the second plate of the shared capacitor, the second transmission gate being configured to close in response to a second input bit being true and to open in response to the second input bit being false.

### Claim 2 (depends on 1)

The compute-in-memory bitcell array of claim 1, wherein the first pair of cross-coupled inverters includes a third output node for a complement of the first stored bit and the first compute-in-memory bitcell further includes a third transmission gate connected between the third output node and the second plate of the shared capacitor, the third transmission gate being configured to open in response to the first input bit being true and to close in response to the first input bit being false.

### Claim 3 (depends on 2)

The compute-in-memory bitcell array of claim 2, wherein the second pair of cross-coupled inverters includes a fourth output node for a complement of the second stored bit and the second compute-in-memory bitcell further includes a fourth transmission gate connected between the fourth output node and the second plate of the shared capacitor, the fourth transmission gate being configured to open in response to the second input bit being true and to close in response to the second input bit being false.

### Claim 4 (depends on 1)

The compute-in-memory bitcell array of claim 1, wherein the first compute-in-memory bitcell further includes:
a first reset n-type metal-oxide semiconductor (NMOS) transistor connected between ground and the second plate of the shared capacitor; and
a first p-type metal-oxide semiconductor (PMOS) transistor connected between the second plate of the shared capacitor and a power supply node for a power supply voltage.

### Claim 5 (depends on 4)

The compute-in-memory bitcell array of claim 4, wherein the second compute-in-memory bitcell further includes:
a second reset NMOS transistor connected between ground and the second plate of the shared capacitor; and
a second PMOS transistor connected between the second plate of the shared capacitor and the power supply node.

### Claim 6 (depends on 5)

The compute-in-memory bitcell array of claim 5, wherein the compute-in-memory array is integrated onto a semiconductor substrate, the first compute-in-memory bitcell being integrated onto a first portion of the semiconductor substrate spanning a first set of five polysilicon lines, the second compute-in-memory bitcell being integrated onto a second portion of the semiconductor substrate spanning a second set of five polysilicon lines.

### Claim 7 (depends on 6)

The compute-in-memory bitcell array of claim 6, further comprising:
a third compute in-memory bitcell that includes a third pair of cross-coupled inverters having a third output node for a third stored bit and includes a third transmission gate connected between the third output node and the second plate of the shared capacitor, the third transmission gate being configured to close in response to a third input bit being true and to open in response to the third input bit being false.

### Claim 8 (depends on 7)

The compute-in-memory bitcell array of claim 7, further comprising:
a fourth compute in-memory bitcell that includes a fourth pair of cross-coupled inverters having a fourth output node for a fourth stored bit and includes a fourth transmission gate connected between the fourth output node and the second plate of the shared capacitor, the fourth transmission gate being configured to close in response to a fourth input bit being true and to open in response to the fourth input bit being false.

### Claim 9 (depends on 8)

The compute-in-memory bitcell array of claim 8, wherein:
the third compute-in-memory bitcell is integrated onto a third portion of the semiconductor substrate spanning a first set of four polysilicon lines and the second compute-in-memory bitcell is integrated onto a fourth portion of the semiconductor substrate spanning a second set of four polysilicon lines.

### Claim 10 (depends on 1)

The compute-in-memory bitcell array of claim 1, wherein the compute-in-memory array is integrated into a machine-learning application.

### Claim 11 (depends on 10)

The compute-in-memory bitcell array of claim 10, wherein the machine-learning application is integrated within a cellular telephone.

### Claim 12 (depends on 1)

The compute-in-memory bitcell array of claim 1, further comprising:
a reset switch connected between the read bit line and a power supply node for a power supply voltage.

### Claim 13 (depends on 12)

The compute-in-memory bitcell array of claim 12, wherein the first compute-in-memory bitcell and the second compute-in-memory bitcell are arranged into a column of the compute-in-memory bitcell array, and wherein the read bit line extends across the column.

### Claim 14 (depends on 1)

The compute-in-memory bitcell array of claim 1, further comprising:
an analog-to-digital converter configured to convert a voltage of the read bit line into a digital word representing a summation of a first multiplication of the first stored bit with the first input bit and a second multiplication of the second stored bit with the second input bit.

### Claim 15 (depends on 14)

The compute-in-memory array of claim 14, wherein the analog-to-digital converter is a successive-approximation-register analog-to-digital converter.

### Claim 16

A compute-in-memory method, comprising:
in a first bitcell, controlling a first pair of transmission gates to drive a second plate of a shared capacitor with a first multiplication signal responsive to a first multiplication of a first input bit with a first stored bit, the shared capacitor having a first plate connected to a read bit line;
sampling a first charge of the read bit line while grounding the second plate of the shared capacitor after the first multiplication to provide a first sampled charge of the read bit line;
in a second bitcell, controlling a second pair of transmission gates to drive the second plate of the shared capacitor with a second multiplication signal responsive to a second multiplication of a second input bit with a second stored bit; and
sampling a second charge of the read bit line while grounding the second plate of the shared capacitor after the second multiplication to provide a second sampled charge of the read bit line.

### Claim 17 (depends on 16)

The compute-in-memory method of claim 16, further comprising:
averaging the first sampled charge of the read bit line with the second sampled charge of the read bit line to provide an averaged charge of the read bit line.

### Claim 18 (depends on 17)

The compute-in-memory method of claim 17, further comprising;
converting the averaged charge of the read bit line into a digital word representing a summation of the first multiplication with the second multiplication.

### Claim 19 (depends on 16)

The compute-in-memory method of claim 16, further comprising:
in a third bitcell, controlling a third pair of transmission gates to drive the second plate of a shared capacitor with a third multiplication signal responsive to a third multiplication of a third input bit with a third stored bit; and
sampling a third charge of the read bit line while grounding the second plate of the shared capacitor after the third multiplication to provide a third sampled charge of the read bit line.

### Claim 20 (depends on 19)

The compute-in-memory method of claim 19, further comprising:
in a fourth bitcell, controlling a fourth pair of transmission gates to drive the second plate of a shared capacitor with a fourth multiplication signal responsive to a fourth multiplication of a fourth input bit with a fourth stored bit; and
sampling a fourth charge of the read bit line while grounding the second plate of the shared capacitor after the fourth multiplication to provide a fourth sampled charge of the read bit line.

### Claim 21 (depends on 16)

The compute-in-memory method of claim 16, further comprising:
performing the first multiplication using an exclusive-not-OR (XNOR) logical operation in the first bitcell.

### Claim 22 (depends on 16)

The compute-in-memory method of claim 16, further comprising:
performing the first multiplication using an exclusive-OR (XOR) logical operation in the first bitcell.

### Claim 23

A compute-in-memory array, comprising:
a read bit line;
a first shared capacitor having a first plate connected to the read bit line;
a second shared capacitor having a first plate connected to the read bit line;
a first set of compute-in-memory bitcells, each compute-in-memory bitcell in the first set of compute-in-memory bitcells having a first capacitor node connected to a second plate of the first shared capacitor; and
a second set of compute-in-memory bitcells, each compute-in-memory bitcell in the second set of compute-in-memory bitcells having a second capacitor node connected to a second plate of the second shared capacitor.

### Claim 24 (depends on 23)

The compute-in-memory array of claim 23, wherein each compute-in-memory bitcell in the first set of compute-in-memory bitcells comprises:
a first inverter cross-coupled with a second inverter;
a first transmission gate connected between a first output node of the first inverter and the first capacitor node; and
a second transmission gate connected between a second output node of the second inverter and the first capacitor node.
