DRAM Comparator Arrays for Low-Latency AI Memory Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current DRAM devices face challenges in efficiently performing large numbers of comparison operations required by machine learning and AI applications, particularly in terms of power consumption and latency, which affects battery life and performance.
Innovation Solution
The development of a semiconductor superlattice structure with enhanced charge carrier mobility, achieved through specific layer configurations and energy band-modifying layers, which reduces conductivity effective mass and improves device mobility, enabling more efficient comparison operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If DRAM devices are used to perform comparison operations for machine learning and AI applications, then the ability to store data values is provided, but power consumption increases and latency is introduced
Solution Approach 1:
The patent combines DRAM storage functionality with comparator functionality into a single integrated device. The DRAM cell structure is merged with comparator circuitry, allowing the same device to both store data values and perform comparison operations, thereby reducing the need for separate storage and processing components that would increase overall power consumption.
Solution Approach 2:
The DRAM device is designed to perform multiple functions: it can store data values in its capacitors and simultaneously perform comparison operations using its integrated comparator arrays. This multi-functionality allows a single device type to handle both memory storage and computational tasks, improving comparison operation efficiency while reducing the power overhead of having separate specialized components.
2Productivity
If DRAM devices are used to perform comparison operations, then data storage is enabled, but latency increases due to the delay between selecting memory locations and data arrival
Solution Approach 1:
By merging the comparator functionality directly into the DRAM cell structure, the patent eliminates the need for data to be fully retrieved from memory before comparison can begin. The comparator arrays are integrated at the memory array level, allowing comparison operations to start while data is still being sensed from the DRAM cells, thereby reducing latency.
3Speed
If superlattice structures are used to enhance charge carrier mobility, then device mobility improves, but device complexity increases due to multiple stacked layers
Solution Approach 1:
The patent modifies the physical and chemical parameters of the semiconductor channel by introducing a superlattice structure with alternating layers of different materials. This changes the energy band structure and effective mass of charge carriers, thereby enhancing mobility. The parameter changes are achieved through controlled deposition of thin films with specific thicknesses and material compositions.
Solution Approach 2:
The superlattice channel is constructed as a composite material structure with alternating layers of different semiconductor or dielectric materials. This composite structure creates a periodic potential landscape that enhances charge carrier mobility through quantum mechanical effects, while the layered composite approach allows for tailored electrical and optical properties.
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 solution enhances the performance of DRAM devices by reducing power consumption and latency, thereby improving the efficiency of comparison operations and extending battery life in applications like machine learning and AI.
Implementation Method 1
an energy band-modifying layer thereon, which may include at least one non-semiconductor monolayer constrained within a crystal lattice of adjacent base semiconductor portions
Implementation Method 2
semiconductor superlattice structure with enhanced charge carrier mobility, achieved through specific layer configurations and energy band-modifying layers, which reduces conductivity effective mass and improves device mobility
Data Source
AI summary
A memory system including a first chip having a processor, and a second chip having a DRAM sector that includes: a plurality of DRAM arrays; an output circuit configured to store a plurality of data values read from the DRAM arrays; a first set of through silicon vias (TSVs) connecting the processor to the DRAM sector, wherein the first processor transmits a plurality of weight data values to the DRAM sector on the first set of TSVs; a plurality of comparator arrays coupled to receive the plurality of weight data values and the plurality of data values read from the DRAM arrays, and in response, generate a plurality of comparison output values; and a second set of TSVs connecting the processor to the DRAM sector, wherein the plurality of comparison output values are transmitted from DRAM sector to the processor on the second set of TSVs.