做人工智慧晶元，最好學學這幾本書

01-27

最近總有同學問我，要是對人工智慧晶元領域感興趣，該怎麼「入門」？我假設大家有計算機或者電子的背景，那麼這四本書應該可以覆蓋所需的知識。

更新：架構方面增加「Computer Organization and Design RISC-V Edition」，相較「Computer Architecture: A Quantitative Approach」，用這本書「入門」可能更好操作一些（特別是RISC-V有很多資源可以參考和實踐）。

演算法

領域的演算法知識是設計專用晶元的基礎，必須掌握。目前深度學習很火，參考資料也非常多。相對來說，這本書雖然比較枯燥，但基本包括了演算法方面必備的知識，可以作為基礎。

1 Introduction
Part I: Applied Math and Machine Learning Basics
2 Linear Algebra
3 Probability and Information Theory
4 Numerical Computation
5 Machine Learning Basics
Part II: Modern Practical Deep Networks
6 Deep Feedforward Networks
7 Regularization for Deep Learning
8 Optimization for Training Deep Models
9 Convolutional Networks
10 Sequence Modeling: Recurrent and Recursive Nets
11 Practical Methodology
12 Applications
Part III: Deep Learning Research
13 Linear Factor Models
14 Autoencoders
15 Representation Learning
16 Structured Probabilistic Models for Deep Learning
17 Monte Carlo Methods
18 Confronting the Partition Function
19 Approximate Inference
20 Deep Generative Models

架構

Computer Abstractions and Technology
Instructions: Language of the Computer
Arithmetic for Computers
The Processor
Parallel Processors from Client to Cloud
Multiprocessors

APPENDIX

The Basics of Logic Design
Graphics and Computing GPUs
Mapping Control to Hardware
Survey of RISC Architectures for Desktop, Server,and Embedded Computers

Computer Design
Chapter 2 Memory Hierarchy Design
Chapter 3 Instruction-Level Parallelism and Its Exploitation
Chapter 4 Data-Level Parallelism in Vector, SIMD, and GPU Architectures
Chapter 5 Thread-Level Parallelism
Chapter 6 Warehouse-Scale Computers to Exploit Request-Level and Data-Level Parallelism
Appendix A Instruction Set Principles
Appendix B Review of Memory Hierarchy
Appendix C Pipelining: Basic and Intermediate Concepts
Appendix D Storage Systems
Appendix E Embedded Systems
Appendix F Interconnection Networks
Appendix G Vector Processors in More Depth
Appendix H Hardware and Software for VLIW and EPIC
Appendix I Large-Scale Multiprocessors and Scientific Applications
Appendix J Computer Arithmetic
Appendix K Survey of Instruction Set Architectures
Appendix L Historical Perspectives and References

這本書不用解釋了，如果沒看過，一定要看；如果你看過，要問自己是不是真正掌握了。

實現

這本書嚴格來說不能算是技術書，但它是了解整個IC產業，以及晶元從設計到實現的完整流程的很好參考。稍有遺憾的是後續沒有更新，一些新的變化可能需要延伸閱讀。

目錄（這本書內容覆蓋面很廣，所以下面列出完整的目錄）：

1 Industry Perspectives
11 Semiconductor Industry
12 Fabless Industry
13 Key Points
2 The Big Picture
21 Electronics Markets
22 The Global Opportunity
23 Some Challenges in Today』s Electronics Marketplace
24 Understanding the Value Chain
25 Customer Needs
26 Overview of Fabless Company Development Activities
27 Key Points
3 Lifecycle of a Fabless IC Company
31 Getting Started
32 Business Plan
33 Funding Process
34 Development Cycle—the Four Phases
35 Roadmap of Products
36 Exit Strategies
37 Long Range Strategies
38 Key Points
4 Selecting the Implementation Approach
41 FPGA, Gate Array ASICs, Semi-custom ASICs
42 If You Had a Fab (IDM)
43 Fabless Sourcing Models (ASSP, COT, ASIC)
44 Design Strategies
45 Key Points
5 Selecting the Technologies
51 Introduction
52 Considerations to Pick the Right Technology
53 Cost per Function (CPF)
54 CPF Reduction from Technology Scaling
55 Die Cost Reduction by Increasing Wafer Size
56 Foundry Financials and Technologies
57 Process Technologies
58 CMOS Challenges
59 The Design Ecosystem
510 Process Alternatives
511 Nanotechnology Co-design Solutions
512 Packaging Consideratons
513 Key Points
6 Implementing the COT Approach
61 Introduction
62 Design Flow and Supply Chain
63 Implementation Time Line
64 Silicon Prototyping and Production
65 Packaging Considerations
66 Test Considerations
67 Quality and Reliability Considerations
68 Supply Chain Considerations
69 Operations Best Practices
610 Operations Effort and Resources
611 Resource Skill Sets
612 Production Operations Activities and Processes
613 Key Points
7 Managing Cost
71 Unit Cost Estimation
72 Optimizing the Die Size and Packing Density per Chip
73 Development Cost
74 Development and Operations Costs
75 Overall Development Cost Estimation
76 Some Cost Tradeoffs
77 Key Points
8 Managing Quality
81 Quality Manual
82 Documentation System
83 Quality in the Development Phase
84 IC Quality and Reliability Qualifi cation
85 Manufacturing Quality
86 Customer Support
87 Key Points
9 Managing the Implementation Program
91 Management at the Vertically Integrated Company
92 Management of the Distributed Supply Chain
93 Comparison of Management Processes
94 Relationship and Partnership Management
95 Program Management
96 Risk Management
97 Design Productivity
98 Key Points
10 Future Trends
101 Industry Stratifi cation and Opportunities
102 Funding
103 Alternatives to the IDM Model
104 Virtual 「Re-Integration」/IFM
105 Order Entry Methodology
106 Managing Innovation
107 The Role of Research Organizations
108 Key Points
Appendices
A1 Business Plan Example
A2 Term Sheet Outline
B1 Transistor Scaling
B2 Yield Models
B3 Quality

編譯

個人感覺區別一個晶元設計者和晶元架構師的差別很大程度在於他對軟體的理解（還包括對產品和產業的理解，即上一本書的內容）。現在很少有不運行軟體的硬體，硬體的效能是由軟體介面體現的。而編譯器作為傳統上硬體和軟體最基本的橋樑，是升級你對硬體的理解應該了解的東西，特別是編譯優化（optimization）和run-time的相關內容。

1 Introduction
2 A Simple Syntax-Directed Translator
3 Lexical Analysis
4 Syntax Analysis
5 Syntax-Directed Translation
6 Intermediate-Code Generation
7 Run-Time Environments
8 Code Generation
9 Machine-Independent Optimizations
10 Instruction-Level· Parallelism
11 Optimizing for Parallelism and Locality
12 Interprocedural Analysis
A A Complete Front End
B Finding Linearly Independent Solutions

除了第一本書之外，其它幾本其實是設計任何晶元都應該熟悉的內容（對編譯的理解算加分項吧）。而每本書涉及的範圍也很廣，真正搞懂並不容易，這一點從大家的評論也可以看出來。但你一旦真搞懂了，你的知識結構基本就沒有什麼死角了。另一方面，這也是你能夠融會貫通，舉一反三的起點。

這裡其實還缺了一個環節，就是晶元設計（把演算法和架構變成可以實現的電路的過程）。比如我之前寫過一本《數字IC設計：方法技巧與實踐》主要覆蓋設計語言，EDA工具等內容。不過這些主要是些技巧和經驗，在實踐中可以很快掌握，需要的時候查一下資料也就可以了。另外，這些內容主要是和數字晶元相關，針對目前比較流行的Deep Learning處理器；基於模擬電路的設計，我不是很熟悉，也就不推薦了。

有了這些背景知識，如果需要快速了解更有針對性的內容，可以看看我關於AI/ML/DL硬體的文章。

T.S.

歡迎關注我的微信公眾號:StarryHeavensAbove

題圖來自網路，版權歸原作者所有