Very High-Speed Integrated Circuit Hardware Description Language, Lecture notes of Verilog and VHDL

Download Very High-Speed Integrated Circuit Hardware Description Language and more Lecture notes Verilog and VHDL in PDF only on Docsity!

Moore’s Law

What happened from 1970 to 2004?

• Transistor count doubles every 24-month
  • Some say every 18-month

Moore’s Law (cont.)

Who’s Moore?

• Gordon Moore a co-founder of Intel Corp.

Corollaries of Moore’s Law

• Performance of computers doubles every two-year
  • Per unit cost
• Power consumption doubles every 18 months
  • Per unit area
• Hard disk capacity doubles every 24 months
• RAM storage capacity doubles every 24 months
• Capital cost of a semiconductor fab doubles every 24 months

More on Moore’s Law on Wiki

Moore’s Law (cont.)

End of Moore’s Law?

Moore’s Law (cont.)

End of Moore’s Law?

• Transistor physical dimensions
  • Gate oxide thickness
• Power wall
  • Amount of power consumed per inch square

What is Reconfigurable Computing?

• RC Definition
  • Study of architectures that can adapt (after

fabrication) to a specific application or

application domain

• Involves architecture, design strategies, CAD

tools, languages, algorithms

• Alternate Definition
  • A way of implementing circuits without

fabricating a device

What is Reconfigurable Computing? (cont.)

• Essentially allows circuits to be implemented as “software” - SW  ISA - RC Circuit ?
• SW: sequence of instructions
• RC Circuit: pattern of RC resources
  • Configured using bitstream
• Circuits no longer same thing as hardware
  • RC devices programmable by downloading bits -

just like software

Why is RC important?

• Tremendous Performance Advantages

• In some cases, 100x faster than uProc
• Similar performances as large cluster
  • But smaller, lower power, cheaper, etc.

Why is RC important? (cont.)

• Example:

• Software executes sequentially
• RC executes all multiplications in parallel
  • Additions become tree of adders
• Even with slower clock, RC is likely much faster
• Performance difference even greater for larger

input sizes

• SW time increases linearly - O(n)
• RC time is basically O(log 2 (n))
  • If enough area is available for (i= 0 ; i < 16 ; i++) y += c[i] * x[i]

Moore’s Law

2010: >1 BILLION transistors!!!! 1993: 1 Million transistors Extremely difficult to design 1BT ➔ ASICs are expensive!

Moore’s Law

2010: >1 BILLION transistors!!!!

• Solution:
  • Make billions of transistors into an RC fabric
  • Fabricate 1 big chip and use it for many things 1993 : 1 Million transistors Make this RC

When to Use RC? (I)

• When it provides cheapest solution
  • Depends on:
    • NRE (Non-Recurring Engineering) cost
      • Cost involved with designing system
    • Unit cost
      • Cost of a manufacturing single device
    • Volume: # of units
• Total cost = NRE + unit cost * volume

When to Use RC? (I)

• uP NRE
  • SW & SW test development
• ASIC NRE
  • Standard cell design
  • Design layout
  • Mask production
  • Test development
• RC NRE
  • HDL & HDL test development
• Unit cost
  • Die cost, yield, testing (?), and packaging

When to Use RC? (I)

• When it provides cheapest solution
  • How about microprocessors?
    • Often provide cheapest solution
      • But not always
    • Often cannot meet performance constraint

Design Process

• 28

• Design Space Exploration

1. Determine architectures that meet requirements

• Performance, power, etc.
• Requires performance analysis/estimation, …

2. Estimate volume of device
3. Determine cheapest solution

• Best Architecture for an Application?

• Cheapest one that meets all design constraints