Buffer Retiming within a Synthesis Environment

IP.com Prior Art Database Disclosure
IP.com Disclosure Number: IPCOM000184450D
Publication Date: 25-Jun-2009
More Like This Download

Publishing Venue

The IP.com Prior Art Database

Abstract

Proposed is a system and method for implementing synthesis retiming for the buffering solutions within our current methodology that does not render the vhdl unreadable to a human. Ultimately, this process allows for synthesis to flip the polarity of a latch within the vhdl with simple updates instead of having to trace through the logic and change net inputs. There are three instances within synthesis that do not produce optimal results. The first occurs when there is a PI feeding into a latch. In this case, design rules dictate that at least one inverter must be present before the input to the latch. But, there are cases where just one inverter causes the polarity of the data to be incorrect so a second inverter must be added which is not optimal. The second case is just like the first case except for a PO. There are many instances where only two inverters would be needed to drive the output of the latch but the polarity of the latch requires a third inverter which adds extra delay to the timing path. The third case where the proposed solution can offer timing improvements is on an internal latch with larger fan-out requiring a buffer tree. The number of buffers in the tree is often dictated by the polarity of the latch or by a specific logic cone in that path and thus, better solutions would be attainable if synthesis was able to control the polarity of the latch. The system uses predefined logic structures that have a variable passed to them controlling the polarity of the signal that is latched. Thus, when synthesis changes the implementation of the logic for retiming, the only update to the vhdl that is required is a parameter override statement beside the latch instantiation to update the vhdl to pass verity. Using this methodology, logic designers can create readable vhdl enables synthesis to implement buffer retimming.

Language

English (United States)

Document File

2 pages / 23.9 KB

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

Buffer Retiming within a Synthesis Environment

Retiming is a method for improving the worst-case timing path by balancing the overall delays between latch stages within a design. This method has been a well-known optimization method for many years however it is not widely automated because of the current methodology that requires a valid, readable vhdl to pass verity and timing paths that stay consistent between each run.

Synthesis is capable of implementing retiming but automated updates to the vhdl create a macro that is incomprehensible for human eyes. Thus, there is not a current system that both offers an automated method of retiming within our current methodology.

The proposed system uses modified entities within the vhdl that are similar to current latch instantiation but with a modified parameter in the generic map statement to enable synthesis access to control the latch polarity. This allows synthesis to implement logic retiming automatically and then update the vhdl directly without having to edit any of the signal names throughout the vhdl file. The parameters can be loaded either within the vhdl itself or in an external include.

Figure 1 below shows a simple example of a timing path which occurs often that can be improved with automatic buffer retiming both an improvement in power and timing.

Figure 1 shows a common implementation for a macro PO. In some cases, three inverters are needed to drive the macro output but in many cases, only two inverters are needed. When that happens, synthesis must choose between either a one inverter solution or a three inverter solution. Thus, if one inverter is not enough to drive the macro PO, an extra inverter is put in the logic path which hurts timing.

The only current solution to thi...

First page image
You are not signed in. If you have an IP.com account, your download price may be lower or waived. Click here if you want to sign-in now.
Loading PayPal...
The full document comprises 2 pages and is available as a PDF document as well as a ZIP archive. The cost is $40.00 USD (depending on your billing address, sales tax may apply); payment may be made directly using your credit card or your PayPal account.

If you've already purchased this document, and wish to download it now you may enter the download access code you received in your original email receipt.