Semiconductor Device Parallel Weight Data Multiplexing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor devices face challenges in reducing processing time while minimizing the increase in occupied area, particularly due to the need to simultaneously read multiple pieces of weight data from local memories and the use of multi-bit width multipliers.
Innovation Solution
A semiconductor device configuration that includes a memory for parallel output of first data, multiple product-sum operation units for these data, and selectors that choose second data based on additional information to perform product-sum operations efficiently, thereby reducing processing time and occupied area.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If multiple pieces of weight data are simultaneously read from local memories and multiple multipliers are used to perform parallel product-sum operations, then processing time is reduced, but the occupied area increases
Solution Approach 1:
The patent combines multiple weight data reads and multiple multiplier operations into a single memory read operation by using a selector to multiplex the weight data output to multiple product-sum operation units. This merging approach allows parallel processing without requiring multiple separate memory accesses or multiple multipliers, thus reducing occupied area while maintaining reduced processing time.
Solution Approach 2:
The selector unit serves multiple functions: it selects which weight data to read from memory, routes the selected weight data to appropriate product-sum operation units, and enables time-division multiplexing of the weight data output. This multi-functionality eliminates the need for dedicated hardware paths for each operation, reducing overall occupied area while preserving parallel processing capability.
2Productivity
If multiple multipliers are used to perform parallel product-sum operations, then processing speed increases, but device complexity increases
Solution Approach 1:
The patent merges multiple multiplier functions into a single multiplier by using a selector to time-division multiplex the weight data output to multiple product-sum operation units. This allows multiple parallel operations to be performed using one shared multiplier resource, thereby increasing processing speed without proportionally increasing device complexity.
Solution Approach 2:
The selector unit dynamically routes the weight data output to different product-sum operation units based on the current operation requirements. This dynamic multiplexing enables the system to perform multiple parallel product-sum operations using a single multiplier, increasing processing speed while keeping device complexity manageable through shared resources.
Data Source
AI summary
A semiconductor device includes: a local memory outputting a plurality of pieces of weight data in parallel; a plurality of product-sum operation units corresponding to the plurality of pieces of weight data; and a plurality of unit selectors corresponding to the product-sum operations units, supplied with a plurality of pieces of input data in parallel, selecting the one piece of input data from the supplied plurality of pieces of input data according to a plurality of pieces of additional information each indicating a position of the input data to be calculated with the corresponding product-sum arithmetic unit calculator in the pieces of input data, and outputting the selected input data. Each of the plurality of product-sum arithmetic units performs a product-sum operation between the weight data different from each other in the plurality of pieces of weight data and the input data outputted from the corresponding unit selector.