micron’s automata processormeseec.ce.rit.edu/551-projects/fall2014/1-1.pdf · microsoft...
TRANSCRIPT
Micron’s Automata Processor
Presented by Jared Mistretta and Ryan Gault
Agenda
� Classical Computing Overview
� Automata Processing Introduction
� Micron’s Automata Processor Architect
� Micron’s Automata Processor Comparisons
� Micron’s Automata Processor Applications
Classical Computing - Architecture
Classical Computing – Execution Model
� F-D-X model
Classical Computing – Execution Model
� Methods for increasing FDX cycle rate� Increasing core clock frequencies� Decreasing memory latency� Pipelining operations� Multi-core and super-scalar architectures
� Problems with these scaling methods� Increases power consumption� Requires more expensive memory architectures� Pipelining is limited scalability� Programming complexity
Classical Computing – Memory Hierarchy
Classical Computing – Memory Hierarchy
Classical Computing – Inherent Problems
� Memory wall
� Sequential operations
� Energy efficiency/ cost of data movement
� Automata processor solves these problems
Automata Processor - Introduction
Automata Processor - Introduction
� Processing In Memory� For large unstructured datasets.� String Matching Machine
Automata Processor - Theory
� Nondeterministic finite state machine
� Ideal characteristics of highly efficient computing systems
� Small operands
� Operator efficiency
� Minimum operand movement
� Result efficiency
� Intrinsic parallelism
Automata Processor – Execution Model
Automata Processor – Computational Elements
� The State Transition Element (STE)
� The Counter Element
� The Boolean Element
Automata Processor – Computational Elements
Common aspects of the elements:1. Each receives sixteen inputs from the routing matrix.
2. Each performs a function within an automaton.
3. Each is programmable.
4. Each originates (drives) one signal back into the routing matrix.
Micron’s Automata Processor
Micron’s Automata Processor – STE
Micron’s Automata Processor –Character Class
Micron’s Automata Processor -Counter Element
Micron’s Automata Processor -Boolean Element
Micron’s Automata Processor -DRAM Comparison
Micron’s Automata Processor –DRAM Comparison
Micron’s Automata Processor –Routing Matrix
Micron’s Automata Processor -Comparisons
Micron’s Automata Processor -Comparisons
� FPGA� Both reconfigurable� Large scale Boolean logic� Control an array of automata processors
� GPU
� Automata threads works independently so the programmer does not have to focus on exploiting parallelism in GPU’s and CPU’s
Micron’s Automata Processor -Applications
� Heterogeneous computing
� Bioinformatics
� Network security
� Signal Intelligence and cryptography
� Finance
� Big Data Domain
Micron’s Automata Processor - Modules
Conclusion
� Micron’s Automata Processor solves complex problems� Solution to big data problems which are becoming more common