Processing-in-Memory Shared I/O Line Architecture

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current memory devices face inefficiencies in processing performance and power consumption due to the need for external communication between processing resources and memory arrays, which can be improved by integrating processing capabilities directly within the memory device.

Innovation Solution

The implementation of a processing-in-memory (PIM) device with a shared input/output (I/O) line that couples sensing circuitry, including a sense amplifier and compute components, directly to the memory array, allowing for simultaneous performance of compute operations with read and write operations within the memory device.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If data is accessed via external bus between processing resources and memory array, then device complexity is reduced, but processing speed decreases and power consumption increases

Engineering Contradiction:
Improveprocessing speedVSAvoiddevice complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The patent merges processing resources (functional units with ALU, FPU, combinatorial logic) directly with the memory array by implementing them on the same chip. This integration eliminates external bus communications, reducing both access time and power consumption while maintaining device functionality.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces an on-chip buffer and sequencer as intermediaries between the memory array and functional units. These components manage data flow and operation sequencing internally, enabling fast processing without requiring external bus transactions.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Use of energy by moving object

If processing resources are implemented externally to memory array, then ease of operation is improved, but power consumption increases

Engineering Contradiction:
Improvepower consumptionVSAvoidease of operation
Core Design Contradiction:
Use of energy by moving objectVSEase of operation

Solution Approach 1:

By combining memory array and processing resources on the same chip, the patent eliminates power-hungry external bus transactions. The integrated architecture performs compute operations directly on stored data, significantly reducing overall power consumption while maintaining operational simplicity through unified control.

Inventive Principle:
Principle #5Merging (Combining)

3Productivity

If compute operations are performed sequentially outside memory, then device complexity is reduced, but productivity decreases

Engineering Contradiction:
Improveprocessing performanceVSAvoiddevice complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent combines multiple functional units (ALU, FPU, combinatorial logic) with the memory array on the same chip, enabling parallel compute operations. This integration allows multiple operations to execute simultaneously on different data sets, dramatically improving processing throughput.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The on-chip buffer pre-loads and sequences data before it reaches the functional units, and the result buffer immediately captures outputs. This preliminary preparation and immediate capture of data eliminate waiting time between operations, enhancing overall processing productivity.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3586334B1Apparatuses and methods for compute in data path
Publication Date: 2023.07.19 MICRON TECHNOLOGY INC
  • EP3586334B1 patent drawingFigure 1A
  • EP3586334B1 patent drawingFigure 1B
  • EP3586334B1 patent drawingFigure 2

AI summary

The present disclosure includes apparatuses and methods for compute in data path. An example apparatus includes an array of memory cells. Sensing circuitry is coupled to the array of memory cells. A shared input/output (I/O) line provides a data path associated with the array. The shared I/O line couples the sensing circuitry to a compute component in the data path of the shared I/O line.