RRAM In-Memory BWT Alignment for DNA Short Reads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
State-of-the-art genome sequencing alignment processes are memory- and compute-intensive, requiring hours or days to align large volumes of short read data due to off-chip bandwidth limitations and inefficiencies, which are not effectively addressed by existing Computing-in-Memory (CIM) technologies.
Innovation Solution
A resistive random access memory (RRAM) based CIM macro chip prototype is designed for Burrows-Wheeler Transformation (BWT) genome sequencing alignment, integrating HfO2 RRAM and 65 nm CMOS to perform XNOR-based match and addition operations, supporting 1- and 2-bit per cell encoding of nucleotides, and leveraging multi-bit properties for improved system parallelism and throughput.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If state-of-the-art alignment processes are used with CPUs/GPUs, then computational accuracy is maintained, but processing time extends to hours or days due to memory-wall limitations
Solution Approach 1:
The patent merges memory and computation into a single integrated RRAM-based computing-in-memory macro. The RRAM crossbar array simultaneously stores reference genome data and performs alignment computations through resistive switching operations, eliminating the memory-wall bottleneck by combining data storage and processing functions into one unified structure.
Solution Approach 2:
The patent replaces traditional CMOS-based mechanical/electronic computing operations with resistive switching operations in RRAM devices. The alignment computations are performed through voltage-driven resistive transitions in the RRAM crossbar array, substituting conventional transistor-based logic operations with more energy-efficient resistive memory operations that enable parallel processing.
2Use of energy by moving object
If conventional CMOS-based computing is used, then device maturity and manufacturing readiness are achieved, but energy efficiency and throughput are limited by off-chip bandwidth
Solution Approach 1:
The patent employs a hybrid composite structure combining RRAM devices with CMOS circuitry. The RRAM crossbar array provides resistive switching for in-memory computation, while integrated CMOS peripheral circuits handle control, sensing, and data processing functions. This composite architecture leverages the complementary strengths of both technologies to achieve high energy efficiency while maintaining manufacturing readiness.
Solution Approach 2:
The patent introduces CMOS peripheral circuits as intermediary components that bridge the RRAM computation core and external systems. These intermediary circuits perform sensing, amplification, and data formatting functions, enabling the RRAM macro to operate independently as a parallel alignment core while maintaining compatibility with standard computing interfaces.
3Productivity
If RRAM-based CIM macro is implemented, then energy efficiency reaches 2.07 TOPS/W with improved throughput, but manufacturing precision requirements increase due to resistive memory variability
Solution Approach 1:
The patent implements sensing amplifiers with feedback mechanisms that read the resistive states of RRAM cells and amplify the differential signals. The feedback loops compensate for resistance variations by comparing reference and data cell states, converting analog resistance values into digital alignment results. This feedback approach mitigates the impact of RRAM variability on computation accuracy.
Solution Approach 2:
The patent employs encoding schemes that map nucleotide sequences to resistive states in the RRAM crossbar array. By transforming biological sequence data into electrical resistance parameters suitable for resistive switching operations, the system leverages the natural parameter compatibility between RRAM devices and genomic data representation, enabling efficient alignment computations despite manufacturing variability.
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
The RRAM-based CIM macro achieves the best energy efficiency with 2.07 TOPS/W and 2.12 G suffixes/J, significantly improving alignment throughput and energy efficiency compared to CPUs/GPUs and prior CMOS ASIC designs.
Implementation Method 1
resistive random access memory (RRAM) based CIM macro chip prototype... HfO2 RRAM... in-memory XNOR-based match circuits
Data Source
AI summary
A system for calculating DNA short-read alignment comprises a memory comprising a plurality of memory cells, the memory comprising first and second regions, a plurality of transmission gates, a plurality of sense amplifiers, each having first and second outputs connected to each bitline in the memory, a plurality of logic gates, each connected to the first and second outputs of each sense amplifier, a digital peripheral circuit connected to the outputs of the logic gates, and a processor configured to perform steps comprising performing a Burrows-Wheeler transform on a nucleotide sequence represented as a string, storing the transformed nucleotide sequence in the first region of the memory, storing a short-read nucleotide sequence in the second region of the memory, activating first and second rows of the memory, and calculating DNA short-read alignment using the digital peripheral circuit.