In-Memory Capacitor Mesh MAC Circuit for Low-Data-Movement Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional computing systems face performance and energy bottlenecks due to data movement between processors and memory, especially in applications requiring large amounts of data, such as machine learning, where in-memory computing can improve processing efficiency.
Innovation Solution
An apparatus comprising a capacitor mesh circuit and a bitcell array, where each bitcell stores a weight bit and multiplies it by an input bit, with the output coupled to a capacitor mesh forming a binary-weighted capacitive voltage divider, enabling partial multiply-accumulate operations and accumulating results in a sample-and-hold circuit for digital conversion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data is stored in separate memory and processed by a processor, then data storage capacity is improved, but data movement overhead and energy consumption increase
Solution Approach 1:
The patent combines memory storage and computational processing into a single integrated structure. Bitcells that store weight values are directly connected to capacitor mesh circuits that perform multiplication and accumulation operations, eliminating the need to move data between separate memory and processor components.
Solution Approach 2:
The capacitor mesh circuit acts as an intermediary between the stored weight data and the computation output. It receives input signals, performs analog multiplication with stored weights through capacitive coupling, and produces MAC results without requiring data to be transferred to a separate processor.
2Adaptability or versatility
If data is moved between processor and memory, then data access flexibility is improved, but processing performance deteriorates
Solution Approach 1:
The computational task is segmented into distributed operations across multiple bitcell columns. Each column independently performs multiplication with its stored weights, and the capacitor mesh circuits aggregate results through parallel signal processing, achieving high performance without centralized data movement.
Solution Approach 2:
The patent transitions from digital sequential processing to analog parallel processing by utilizing voltage and capacitance dimensions. Weight values are stored as capacitance values in the capacitor mesh, enabling simultaneous multiplication operations across multiple data points in a single computational cycle.
3Productivity
If in-memory computing is implemented, then processing performance is improved, but device complexity increases
Solution Approach 1:
The bitcell structure serves multiple functions: it stores weight values in its capacitive elements, receives input signals, performs multiplication operations, and outputs results to the capacitor mesh. This multi-functionality reduces the need for separate dedicated components for each operation.
Solution Approach 2:
The capacitor mesh circuit automatically performs the multiplication and accumulation operations when input signals are applied. The stored weight capacitances interact with input voltages through fundamental electrical principles, producing MAC results without requiring external control logic or additional processing steps.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances processing performance by reducing data movement and energy consumption through in-memory computations, facilitating efficient multiply-accumulate operations in digital signal processing and machine learning algorithms.
Implementation Method 1
each signal line in the plurality of signal lines is electrically coupled such that the capacitor mesh circuit forms a binary-weighted capacitive voltage divider circuit between the plurality of signal lines
Implementation Method 2
a multiplication unit configured to multiply the weight bit by an input bit of an input value and to provide a result of the multiplication to an output of the bitcell
Data Source
AI summary
According to an example embodiment, an apparatus comprises a capacitor mesh circuit comprising a plurality of signal lines and a bitcell array comprising a plurality of bitcells.