Semiconductor Device Neural Network Processing Decompressor Switch Circuit

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing semiconductor devices for neural network processing face challenges in efficiently utilizing accumulators and managing large weighting factor data, leading to potential increases in circuit area and power consumption when using decompressors either between memory and DMA controllers or between DMA controllers and accumulators.

Innovation Solution

A semiconductor device configuration that includes a decompressor, a switch circuit between the decompressor and the accumulator unit, and a DMA controller, which restores compressed weighting factors and transfers them to multiple accumulators based on identifiers, allowing for efficient utilization of resources and reducing circuit area and power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a decompressor is placed between the memory and the DMA controller, then the weighting factor data can be restored before transfer, but the accumulators cannot be effectively utilized and the processing speed decreases

Engineering Contradiction:
Improvedata restoration accuracyVSAvoidprocessing speed
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The decompressor performs preliminary restoration of weighting factor data in advance, preparing the uncompressed data before the actual neural network processing begins. This allows the accumulators to receive ready-to-use data and operate at full capacity during the processing phase, resolving the contradiction between data restoration reliability and processing speed.

Inventive Principle:
Principle #10Preliminary action

2Productivity

If a decompressor is placed between the DMA controller and the accumulators, then the accumulators can utilize the data effectively, but a compressor is needed for each accumulator which increases circuit area and power consumption

Engineering Contradiction:
Improveaccumulator utilization efficiencyVSAvoidcircuit area
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

A single decompressor unit is designed to serve multiple accumulators simultaneously. The decompressor restores weighting factor data once, and the restored data is then distributed to all required accumulators through the switch circuit. This universal approach eliminates the need for separate compressors at each accumulator, reducing circuit area and power consumption while maintaining high accumulator utilization.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Productivity

If a decompressor is placed between the DMA controller and the accumulators, then the accumulators can utilize the data effectively, but it increases power consumption due to requiring a compressor for each accumulator

Engineering Contradiction:
Improveaccumulator utilization efficiencyVSAvoidpower consumption
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The single shared decompressor serves all accumulators, eliminating redundant compressor circuits. By consolidating the decompression function into one universal unit that serves multiple accumulators, the system achieves high accumulator utilization while minimizing power consumption through resource sharing and elimination of duplicate hardware.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Quantity of substance

If compressed weighting factor data is stored in memory, then the memory usage is reduced, but additional decompression processing is required which increases device complexity

Engineering Contradiction:
Improvememory storage capacityVSAvoiddecompression processing complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The decompressor acts as an intermediary component between the compressed data in memory and the accumulators that require uncompressed data. This mediator performs the necessary decompression transformation, enabling the system to store data efficiently in compressed form while providing uncompressed data to the processing units through a dedicated intermediate device.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS20220398441A1Semiconductor device
Publication Date: 2022.12.15 RENESAS ELECTRONICS CORP
  • US20220398441A1 patent drawing
  • US20220398441A1 patent drawing
  • US20220398441A1 patent drawing

AI summary

A semiconductor device executes the processing of a neural network. The memory MEM1 holds a plurality of pixel values and j compressed weighting factors. The decompressor DCMP restores the j compressed weighting factors to the uncompressed k (k≥j) weighting factors. The DMA controller DMAC1 reads the j compressed weighting factors from the memory MEM1 and transfers them to the decompressor DCMP. The n (n>k) accumulators in the accumulator unit ACCU multiply a plurality of pixel values and k uncompressed weighting factor to accumulate and add the multiplication results to the time series. A switch circuit SW1 provided between the decompressor DCMP and the accumulator unit ACCU transfers the k uncompressed weighting factors restored by the decompressor DCMP to n accumulators based on the correspondence represented by the identifier.