Замыкание дизайна в проектировании ВЛСИ: от концепции к производству.

Stage of VLSI semiconductor design workflow

Design Closure is a part of the digital electronic design automation workflow by which an integrated circuit (i. e. VLSI) design is modified from its initial description to meet a growing list of design constraints and objectives. Every step in the IC design (such as static timing analysis, placement, routing, and so on) is already complex and often forms its own field of study. This article, however, looks at the overall design closure process, which takes a chip from its initial design state to the final form in which all of its design constraints are met.

Every chip starts off as someone’s idea of a good thing: "If we can make a part that performs function X, we will all be rich!" Once the concept is established, someone from marketing says "To make this chip profitably, it must cost $C and run at frequency F." Someone from manufacturing says "To meet this chip’s targets, it must have a yield of Y%." Someone from packaging says “It must fit in the P package and dissipate no more than W watts.” Eventually, the team generates an extensive list of all the constraints and objectives they must meet to manufacture a product that can be sold profitably. The management then forms a design team, which consists of chip architects, logic designers, functional verification engineers, physical designers, and timing engineers, and assigns them to create a chip to the specifications.

The distinction between constraints and objectives is straightforward: a constraint is a design target that must be met for the design to be successful. For example, a chip may be required to run at a specific frequency so it can interface with other components in a system. In contrast, an objective is a design target where more
(or less) is better. For example, yield is generally an objective, which is maximized to lower manufacturing cost. For the purposes of design closure, the distinction between constraints and objectives is not important; this article uses the words interchangeably.

Designing a chip used to be a much simpler task. In the early days of VLSI, a chip consisted of a few thousand logic circuits that performed a simple function at speeds of a few MHz. Design closure was simple: if all of the necessary circuits and wires "fit", the chip would perform the desired function. Modern design closure has grown orders of magnitude more complex. Modern logic chips can have tens to hundreds of millions of logic elements switching at speeds of several
GHz. This improvement has been driven by Moore’s law of scaling of technology, and has introduced many new design considerations. As a result, a modern VLSI designer must consider the performance of a chip against a list of dozens of design constraints and objectives including performance, power, signal integrity, reliability, and yield. In response to this growing list of constraints, the design closure flow has evolved from a simple linear list of tasks to a very complex, highly iterative flow such as the following simplified ASIC design flow:

Concept phase: Functional objectives and architecture of a chip are developed. Logic design: Architecture is implemented in a register transfer level (RTL) language, then simulated to verify that it performs the desired functions. This includes functional verification. Floorplanning: The RTL of the chip is assigned to gross regions of the chip, input/output (I/O) pins are assigned and large objects (arrays, cores, etc.) are placed. Logic synthesis: The RTL is mapped into a gate level netlist in the target technology of the chip. Design for Testability: The test structures like scan chains are inserted. Placement: The gates in the netlist are assigned to nonoverlapping locations on the chip. Logic/placement refinement: Iterative logical and placement transformations to close performance and power constraints. Clock insertion: Balanced buffered clock trees are introduced into the design. Routing: The wires that connect the gates in the netlist are added. Postwiring optimization: Remaining performance, noise, and yield violations are removed. Design for manufacturability: The design is modified, where possible, to make it as easy as possible to produce. Signoff checks: Since errors are expensive, time consuming and hard to spot, extensive error checking is the rule, making sure the mapping to logic was done correctly, and checking that the manufacturing rules were followed faithfully. Tapeout and mask generation: the design data is turned into photomasks in mask data preparation.