Browse Prior Art Database

Temporarily use multiple power rails for increased integrator flexibility

IP.com Disclosure Number: IPCOM000124405D
Original Publication Date: 2005-Apr-19
Included in the Prior Art Database: 2005-Apr-19
Document File: 5 page(s) / 63K

Publishing Venue

IBM

Abstract

Disclosed is a system to allow for increased integrator flexibility in floorplannable object placement on a densely wired Integrated Circuit chip by adding multiple power rails which can be deleted later if needed. This answers the key design question of where the higher metal layer power rails should be placed to reduce unused gaps between floorplannable objects, and hence the overall chip height and/or width, which is often critical.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 45% of the total text.

Page 1 of 5

Temporarily use multiple power rails for increased integrator flexibility

The permute pipe in the Vector-Scalar-Unit (VSU) is densely wired, needing to use M1,M2,M3, M4 and even M5 metal layers internally in several pipe floorplannable object layouts, where M1 is the metal layer closest to the silicon, M2 is further and so on. On this particular project, the M1,M2,M3 power rails did not have to connect to any unit level power rails on these metal layers. On this project, M4 is a horizontal metal layer, M5 is a vertical metal layer. There was a requirement for a periodic power grid on both M4 and, if used, M5 that connected to the integrator's power grids on the same metal layers. The permute pipe's placement in the horizontal direction was not critical, the integrator had several extra tracks of available space. So the choice made for M4 power rails acceptably snapped to the unit's M4 power rails and there was no need to give the integrator increased flexibility on M4.

     However, on the M5 metal layer, the permute pipe had placement uncertainty since many pipe macros (floorplannable objects) needed to use M5, and thus needed to have pre-set M5 power rails prior to their internal wiring and before ALL layouts would be done and ALL floorplannable object heights would be known precisely. The key design question is: Where should the M5 power rails be put to reduce unused gaps between the floorplannable objects, and hence the overall pipe height, which was very critical? This idea of using multiple power rails in advance solves, or at least reduces that problem. Although the multiple power rails can be left in the macro for additional power robustness (if the methodology checkers are adjusted to accept this), removing them is also perfectly valid if additional wiring tracks are useful to the macro signal wires. So the end "cost" of this approach is zero additional wiring resource.

     First done with M5, but applicable to any metal layer, this idea uses multiple power rails within a macro to allow integrators increased flexibility in placement. An ability to switch VDD and GND net designations in the macro gives the integrator additional flexibility. This approach is useful for densely wired pipes with fewer unit wires going overhead. It is also useful where dimension (either height, width, or both) is particularly critical. This approach is to essentially add multiple interleaving power grids in the macro as internal macro wiring permits. The unused power rails can be removed later or can be kept for additional robustness if macro and unit wiring allows.

     The normal methodology is to add internal macro M5 power rails every 16 tracks (16 tracks = 1 bit pitch in this methodology). Power rails are best added before wiring the signal portion of the macro. Macro 1 may be wired before Macro 2's final layout height is known. Perhaps Macro 1 is based on logic reuse while Macro 2 is entirely new. In this case, Macro 2 will probably not even have a co...