Accelerating AI/ML Applications with New NVM (1 Day)
$695 – NAND flash non-volatile memory (NVM) took over 30 years to emerge as a broad-based data storage alternative. But now that flash has established the path, we can expect other new NVM technologies to emerge more quickly. One key is matching each technology’s unique characteristics to applications that can benefit from them the most. Many of these are artificial intelligence and machine learning (AI/ML) applications.
This seminar examines the fundamental calculations at the heart of most AI/ML algorithms. These include scaling, offsets, additions, inner products, multiply accumulates, and matrix calculations for convolution. It will explain the fundamentals of emerging NVM technologies and how they can be applied to improve the performance of training and inference hardware, and will comprehensively cover AI/ML algorithms, architectures, and frameworks.
The most popular NVM technologies are introduced and their principles of operation, their pros and cons, and what new developments to expect are explained. You’ll learn the fundamentals of phase change memory, 3D XPoint™, ReRAM, memristor, spin-transfer torque MRAM, and other emerging technologies, as well as how they are being (or may be) applied to AI/ML hardware.
Many new NVM technologies use sophisticated concepts from quantum physics. Your instructor, Chuck Sobey, is well-known for his ability to make difficult technological concepts clear and useful. You’ll be able to ask this veteran storage R&D consultant your most pressing questions and draw on his technical expertise and industry insight.
Reliable, unbiased information is critical for making the right decisions! So don’t proceed (and spend a lot of development money) without a clear understanding of the fundamentals of these rapidly-evolving technologies, and where they can make the biggest impact for your company’s AI/ML strategy! This seminar will quickly get you up-to-speed with the important new NVM technologies.
This tutorial is designed for engineers, managers, and executives who need to make immediate decisions. It is presented by KnowledgeTek, the world’s leading data storage technology training company. Suggested prerequisites include a technical background and interest in data storage.
Link to PDF of coursebook for FMS 2019 is below:
|1. FUNDAMENTALS OF AI/ML|
• Big data, training, inference, and accelerators
• Von Neumann architecture
• Overview of CNNs and deep learning
• AI/ML math
• Identifying the bottlenecks
2. THE MEMORY HIERARCHY
• SRAM and caches
• NAND and NOR flash
• The device determines the system
• Storage and persistent memory (PM)
• Why do we need new NVM technologies?
3. CHANGING THE HIERARCHY
• High speed 2.5D memory options
• Computational storage
• Numerical precision options
• Analog vs. digital
• Neuromorphic computing
4. COMPETING (OR CO-EXISTING) TECHNOLOGIES
• Array architectures
• Access devices
• Phase change memory (PCM, PC-RAM)
• 3D XPoint and QuantX
• RRAM: Filamentary, interface, bulk, ion transport, memristor
• MRAM: Spin-transfer torque (STT-RAM)
• Other emerging NVMs
5. SYSTEM-LEVEL CONSIDERATIONS
• Latency and bandwidth
• Errors and signal processing
• Endurance and retention
• Error correction coding (ECC)
6. NEXT STEPS TO ACCELERATE AI/ML BENEFITS
• Gain experience with emerging memories
• AI/ML state-of-the-art is still changing rapidly
• Match the technology to the application
• Additional NMV technologies to watch
• The device determines the system: Revisited