Intel® Quartus® Prime Standard Edition User Guide: Design Recommendations

ID 683323
Date 9/24/2018
Public
Download
Document Table of Contents

1.2.1.2. Avoid Unintended Latch Inference

Avoid using latches to ensure that you can completely analyze the timing performance and reliability of your design. A latch is a small circuit with combinational feedback that holds a value until a new value is assigned. You can implement latches with the Intel® Quartus® Prime Text Editor or Block Editor.

A common mistake in HDL code is unintended latch inference; Intel® Quartus® Prime Synthesis issues a warning message if this occurs. Unlike other technologies, a latch in FPGA architecture is not significantly smaller than a register. However, the architecture is not optimized for latch implementation and latches generally have slower timing performance compared to equivalent registered circuitry.

Latches have a transparent mode in which data flows continuously from input to output. A positive latch is in transparent mode when the enable signal is high (low for a negative latch). In transparent mode, glitches on the input can pass through to the output because of the direct path created. This presents significant complexity for timing analysis. Typical latch schemes use multiple enable phases to prevent long transparent paths from occurring. However, timing analysis cannot identify these safe applications.

The Timing Analyzer analyzes latches as synchronous elements clocked on the falling edge of the positive latch signal by default. It allows you to treat latches as having nontransparent start and end points. Be aware that even an instantaneous transition through transparent mode can lead to glitch propagation. The Timing Analyzer cannot perform cycle-borrowing analysis.

Due to various timing complexities, latches have limited support in formal verification tools. Therefore, you should not rely on formal verification for a design that includes latches.