# US 11,029,745 B2 · Systems and methods for controlling instantaneous current changes in parallel processors

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

- Number: US 11,029,745 B2
- Document ID: US11029745B2
- Status: Granted
- Kind: B2
- Application number: 16/184,934
- Filed: 2018-11-08
- Published: 2020-03-05
- Granted: 2021-06-08
- Earliest claimed: 2018-08-31
- Assignees: QUALCOMM Incorporated
- Applicants: Qualcomm Incorporated
- Inventors: Kyle Ernewein; Jason Edward Podaima; Francisco Perez; John Daniels; Alex Miler; Jeffrey Gemar; Rexford Alan Hill; Haoping Xu
- Web page: https://patentgrep.com/patent/US11029745B2

## CPC

- G06F 1/324: by lowering clock frequency
- G06F 1/3206: Monitoring of events, devices or parameters that trigger a change in power modality
- G06F 1/3228: Monitoring task completion, e.g. by use of idle timers, stop commands or wait commands
- Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE

## Abstract

Systems and methods are disclosed method for controlling instantaneous current changes in parallel processors with arrays of parallel computing elements, such as neural processors. An exemplary method comprises monitoring the array of computing elements and determining a transition from a first activity level of the array to a second activity level of the array, such as an idle-to-active or active-to-idle transition. Once a transition is determined, the array is selectively controlled to minimize the instantaneous current change from the transition from the first activity level to the second activity level.

## Claims

### Claim 1

A method for controlling an instantaneous current change in a parallel processor, the method comprising:
monitoring an array of computing elements of the processor, the computing elements configured to operate independently of each other in parallel;
determining a transition from a first activity level of the array to a second activity level of the array; and
selectively controlling the array to minimize the instantaneous current change from the transition from the first activity level to the second activity level; wherein selectively controlling the array comprises controlling a frequency of a core clock for the processor; and controlling the frequency of the core clock comprises first decreasing the frequency of the core clock and then incrementally increasing the frequency of the core clock; and the first activity level is an idle state and the second activity level is an active state.

### Claim 2 (depends on 1)

The method of claim 1, wherein:
determining the transition comprises predicting a future change from the first activity level to the second activity level.

### Claim 3 (depends on 1)

The method of claim 1, wherein:
determining the transition comprises determining a present change from the first activity level to the second activity level;
the array comprises a plurality of portions, each portion comprising a set of the computing elements; and
selectively controlling the array comprises incrementally transitioning each of the plurality of portions of the array from the first activity level to the second activity level until all of the plurality of portions are in the second activity level.

### Claim 4 (depends on 3)

The method of claim 3, wherein:
transitioning each of the plurality of portions of the array from the first activity level to the second activity level comprises providing operands to the computing elements of one or more of the plurality of portions of the array, the operands causing the computing elements to execute.

### Claim 5 (depends on 4)

The method of claim 4, wherein the operands causing the computing elements to execute are dummy operands for which no output is recorded, the method further comprising:
once all of the plurality of portions of the array are in the second activity level, providing real operands to all of computing elements of the array.

### Claim 6 (depends on 4)

The method of claim 4, wherein the operands causing the computing elements to execute are real operands, the method further comprising:
delaying the real operands for the portions of the array which have not transitioned to the second activity level; and
re-aligning the timing of the outputs of the computing elements of the portions of the array.

### Claim 7

A system for controlling an instantaneous current change in a parallel processor, the system comprising:
an array of computing elements of the processor, the computing elements configured to operate independently of each other in parallel;
a controller configured to determine a transition from a first activity level of the array to a second activity level of the array; and
the controller configured to control the array to minimize the instantaneous current change from the transition from the first activity level to the second activity level; the controller further comprises a clock rate controller configured to control the frequency of a core clock of the processor; the clock rate controller is further configured to first decrease the frequency of the core clock and then incrementally increase the frequency of the core clock, wherein the first activity level is an idle state and the second activity level is an active state.

### Claim 8 (depends on 7)

The system of claim 7, wherein:
the controller comprises an activity monitor configured to predict a future change from the first activity level to the second activity level.

### Claim 9 (depends on 7)

The system of claim 7, wherein:
the controller is configured to determine the transition by determining a present change from the first activity level to the second activity level;
the array comprises a plurality of portions, each portion comprising a set of the computing elements; and
the controller is configured to incrementally transition each of the plurality of portions of the array from the first activity level to the second activity level until all of the plurality of portions are in the second activity level.

### Claim 10 (depends on 9)

The system of claim 9, wherein:
the controller is further configured to incrementally transition each of the plurality of portions of the array from the first activity level to the second activity level by providing operands to the computing elements of one or more of the plurality of portions of the array, the operands causing the computing elements to execute.

### Claim 11 (depends on 10)

The system of claim 10, wherein:
the operands causing the computing elements to execute are dummy operands for which no output is recorded; and
the controller is further configured to cause real operands to be provided to all of the computing elements of the array once all of the portions of the array are in the second activity level.

### Claim 12 (depends on 10)

The system of claim 10, wherein the operands causing the computing elements to execute are real operands, the system further comprising:
one or more delay elements configured to delay providing the real operands for the portions of the array which have not transitioned to the second activity level; and
an output element configured to re-align the timing of the outputs of the computing elements of the portions of the array while portions of the array have not transitioned to the second activity level.

### Claim 13 (depends on 7)

The system of claim 7, wherein the parallel processor is a neural processor.

### Claim 14

A system for controlling an instantaneous current change in a parallel processor, the system comprising:
an array of computing elements of the processor, the computing elements configured to operate independently of each other in parallel;
a local memory coupled to the array of computing elements, the local memory configured to buffer a stream of operands provided to the array of computing elements; the local memory is configured to receive a plurality of operands in an order different from an order in which the local memory provides the plurality of operands to the array of computing elements; and
a controller configured to determine, based on movement of the operands through the local memory, a transition from a first activity level of the array to a second activity level of the array; the controller is configured to control the array to minimize the instantaneous current change from the transition from the first activity level to the second activity level, the controller further comprises: an activity monitor which detects the first activity and second activity levels of the array, and a clock rate controller; the clock rate controller configured to control the frequency of a core clock of the processor.

### Claim 15 (depends on 14)

The system of claim 14, wherein:
the local memory comprises an input first-in first-out (FIFO) buffer configured to store the operands; and
the activity monitor is configured to predict a future change from the first activity level to the second activity level based on a fill level of the input FIFO buffer.

### Claim 16 (depends on 15)

The system of claim 15, wherein:
the first activity level is an active state and the second activity level is an idle state; and
the activity monitor is configured to cause the clock rate controller to incrementally decrease the frequency of the core clock in response to a prediction of a transition from the active state to the idle state.

### Claim 17 (depends on 16)

The system of claim 16, further comprising an output FIFO buffer configured to store results provided by the array of computing elements, wherein the activity monitor is configured to predict a future change from the active state to the idle state when a fill level of the input FIFO buffer decreases below a first threshold or when a fill level of the output FIFO buffer increases above a second threshold.

### Claim 18 (depends on 14)

The system of claim 14, wherein:
the activity monitor is configured to predict a future change from the first activity level to the second activity level based on a comparison of a location in a reference system having three or more indices of a current operand provided to the array with a threshold in the reference system.

### Claim 19 (depends on 18)

The system of claim 18, wherein:
the first activity level is an active state and the second activity level is an idle state; and
the activity monitor is configured to cause the clock rate controller to incrementally decrease the frequency of the core clock in response to a prediction of a transition from the active state to the idle state.

### Claim 20 (depends on 19)

The system of claim 19, wherein the activity monitor is configured to predict a future change from the active state to the idle state when the comparison indicates a current operand block in the local memory is within a threshold of being completely provided to the array and a next operand block is not present in the local memory.

### Claim 21 (depends on 14)

The system of claim 14, wherein the parallel processor is a neural processor.