Version Nichts ist wahr. Alles ist erlaubt.
Lecture: Place & route on silicon
A gentle introduction to place & route algorithms for digital integrated circuits
After a brief introduction to digital circuits this talk will outline placement and routing algorithms used for creating digital integrated circuits.
This talk introduces the algorithms used for placement and routing of digital integrated circuits.
The talk does *not* cover:
* high-level circuit design (The art of creating meaningful circuits. Often done with languages like Verilog, VHDL, SpinalHDL, Chisel, Amaranth, etc )
* logic synthesis (Converts the high-level description into a graph-like circuit description, called netlist)
Place-and-route refers to the transformation of a graph-like circuit description (netlist) into a geometrical representation of the circuit (layout).
The netlist is typically produced by logic synthesis. The netlist consists of many sub-circuits, so called "standard-cells" but also "macro cells".
Standard-cells implement simple logic functions such as inverters, logical "and", "nand", "xor", and storage elements.
The netlist may also import larger pre-compiled macro cells such as SRAM blocks.
For a physical implementation of the circuit, the sub-circuits need to be placed on the chip surface and need to be connected (routed) using metal wires.
Transforming the netlist into a layout typically requires the following input data:
* A netlist of the circuit, of course.
* A set of constraints: For example the desired clock frequency and area of the circuit.
* Design rules: A set of constraints required for successful fabrication. This typically involves geometrical constraints such as minimum width and spacing of metal wires.
* A standard-cell library: This is a set of building-blocks usually used to assemble the circuit. The library contains the geometrical layout of the standard-cells and also information about their timing behavior.
Then the following steps convert the input data into a layout:
* IO-planning: Decide where to put the input and output pins of the circuit.
* Floor-planning: Decide how to geometrically arrange various parts of a larger system.
* Power distribution: Insert regular rows of metallic power-rails which supply the standard-cells with energy
* Global placement: Decide where to roughly place the standard-cells such that the wiring will short and possible
* Tie-cell insertion: Provide constant 0 and 1 signals, where needed.
* Clock-tree synthesis: Storage elements typically need a clock-signal. Often the clock signal needs to be distributed to a large number of storage elements.
* Detail placement: Do fine-tuning, such as snapping the standard-cells to a grid
the signal propagation delay from the clock source to the storage elements should be more-or-less equally distributed.
* Optimizations to meet timing requirements: Some signals might be too slow or to fast. There's a variety of techniques to improve this, such as amplifying signals with buffers.
* Routing: The placed cells need to be connected with metal wires.
* Filler insertion: fill unused space for example with capacitors to stabilize the supply voltage
* Verification: Make sure all constraints are met. Otherwise, try to fix the circuit and repeat above steps in order to converge to a valid solution.
This talk will focus on a widely used algorithm for global placement and introduces basic principles of routing algorithms.