Hastlayer is a tool that allows computer programs written in languages like C# and Python to be converted into logic that runs on Field Programmable Gate Arrays (FPGAs), providing performance and efficiency benefits. This is done by translating the program's logic into a format that can be executed by the FPGA's programmable circuitry. Potential uses include artificial intelligence, image processing, and cryptography. The presentation demonstrated converting a program using a new binary format called posits that requires less space than traditional floating point numbers.

1. Turning software into
computer chips - Hastlayer
Zoltán Lehóczky and Benedek Farkas @ Lombiq
J on the Beach May 24th, 2018

2. 2

3. 3

4. Warning
Hastlayer is currently in alpha stage!
4

5. What’s Hastlayer?
5

6. computer program → computer chip (hardware)
6

7. logic expressed as software → logic expressed as hardware
7

8. …with FPGAs
• Field-Programmable Gate Array
• Can behave like any other chip (with limitations)
• Can dynamically be „re-wired”
8
Image by SparkFun Electronics, Boulder, USA

9. FPGAs?
• If you use Bing or Azure, you’ve used them!
• Found in routers, X-ray machines, self-driving cars…
• You need to be a hardware engineer to utilize them
9

10. So is it the same as a…
• Raspberry Pi: no, that’s a computer
• Arduiono/Netduino: no, those are microcontrollers
10

11. computer program → FPGA logic
11

12. .NET (C#, VB, C++, F#, Python, PHP, JavaScript…) → FPGA logic
12

13. What’s the point?
13

14. CPU vs GPU vs FPGA
14
Parallelism Program complexity Power efficiency
CPU GPU FPGA

15. But!
15
CPU GPU FPGA FPGA with Hastlayer
How hard to learn?

16. The benefits of FPGAs for us all
• Performance increase for parallel compute-bound algorithms
• Higher power efficiency
• Still only software development
16

17. Where would I use FPGAs?
• Artificial intelligence, machine learning
• Image and video processing, computer vision
• Algorithmic trading
• Data compression
• Cryptography and cryptocurrencies
• Embedded sytems (robots, IoT)
17

18. Demo: Hands-on Hastlayer
18

19. Next-Generation Arithmetics
19

20. Why should I care?
20

21. Problems with IEEE floats
• Wasteful – over 16 million bit patterns mean NaN
• (but none of them says Batman)
• Special cases require complex logic
• Value distribution is not very user-friendly
• Dynamic range and precision is locked
21

22. How does a posit look like?
Source: posithub.org

23. Posit construction
23Source: posithub.org

24. Posit positives
• Less bits to store the same information
• Also affects energy efficiency
• Each value uses the same logic → simpler HW
• No CPU flag black magic
• Flexibility through configurable exponent size
• Consistent results across platforms
24

25. Distribution of values – posits vs floats
Source: posithub.org

26. Where can I get it and use it?
• FOSS implementations in multiple languages
• .NET C# implementation by Lombiq
• Transformed to FPGA using Hastlayer
26

27. Demo: 32-bit posits in action
27

28. 28

29. Next steps
• More operations (division, exponentiation, trigonometric functions,
fused operations, etc.)
• 8, 16, 64 bit versions with code generation
• Reduced-precision for DNN
• Valids (interval arithmetics)
29

30. Wrapping up
30

31. And more!
• More use-cases
• HPC-ready
• FPGAs in every datacenter (AWS, Azure, Intel)
31

32. I like this, how do I start?
• Check out the SDK: https://github.com/Lombiq/Hastlayer-SDK/
• Be ready for an FPGA-filled future!
32

33. Are you ready to *be* the hardware?
• crew@hastlayer.com
• https://hastlayer.com
• https://hastlayer.com/arithmetics
• https://github.com/Lombiq/Hastlayer-SDK/
33