How to Design Analog/Mixed Signal (AMS) at 28nm

Wireless, networking, storage, computing and FPGA applications have been moving aggressively to advanced process nodes to take advantage of lower power consumption, improved performance and area reduction. Today, most of these applications integrate a significant amount of analog/mixed signal (AMS) or RF together with digital circuits. Since AMS often occupies over 50% of the chip area, applying traditional, conservative approaches when migrating to an advanced node diminishes and possibly eliminates these benefits.

Due to significant changes in physical effects and device performance, a simple migration to next node is not practical. AMS circuits need to be optimized and often completely redesigned to meet performance specs. This requires design companies to have an AMS IP flow fully ready at the same time as, or even earlier than, the digital flow in order to realize silicon at advanced process nodes.

A survey of 561 predominantly analog and mixed-signal designers and CAD engineers from over 150 companies, collected during Cadence worldwide Mixed-Signal Seminars in March 2011, confirmed that 65nm has became mainstream for mixed-signal. The survey also showed strong AMS design activity at 40 and 28nm.

 

Figure 1 - Analog/mixed-signal designers look to lower process nodes

Challenges of Advanced Process Nodes

Designing AMS chips at advanced nodes increases some of existing challenges and brings new challenges. The main challenges include parametric variation, device reliability, layout dependent effects (LDE) and overall design productivity.

Traditionally, designers analyze the impact of parametric variations and use more robust circuit topologies and statistical circuit optimization techniques to center designs to meet specifications with acceptable yield. Advanced node analog circuits increasingly require self-calibration to cope with increased parametric variation and process drift. Self-calibration typically involves digital logic implemented tightly with analog circuits, requiring a mixed-signal flow instead of pure analog design flow.

At advanced nodes, devices are more susceptible to effects like Time Dependent Dielectric Breakdown (TDDB), Hot Carrier Aging (HCA) and Negative Bias Temperature Instability (NBTI). Any of these can cause device failure, particularly if over-voltage conditions persist over longer periods. Therefore it is very important to identify devices with over-voltage conditions across all modes of operation and perform reliability analysis early in the design phase.

Due to LDE, device current and threshold voltage vary depending on the surrounding layout. This is mainly caused by well proximity and stress effects, and variation could be 10% or even more. To avoid late layout re-work and schedule delays, it is important to understand impact of LDE on circuit performance, identify the most sensitive devices, and implement them in silicon with special care keeping variation within acceptable tolerances.

Increased challenges require a higher level of design automation to keep up design productivity at advanced process nodes. Accurately predicting circuit performance in pre-layout stages, analyzing sensitivity and identifying the most critical devices, capturing design intent, communicating intent as a constraint to layout designers, and giving designers the means to construct layouts correctly are all becoming mandatory.

TSMC AMS Reference Flow

Through the AMS reference flow initiative, Cadence and TSMC are closely collaborating in addressing challenges of AMS design for the 28nm process. As result of this collaboration, AMS v1.0 was released last year, followed by AMS v2.0 released earlier this month. AMS v2.0 adds key features including LDE-aware layout, advanced Monte Carlo Analysis, device reliability analysis, Analog Base Sub-circuits (ABS) optimization and Multi Technology Simulation (MTS) for 3D-IC/package integration.

In AMS v2.0, the LDE calculation engine is tightly integrated in the Virtuoso environment to provide almost instantaneous feedback to layout designers on the quality of device placement through comparison of Vth and Id_sat variations of the particular device against specified constraints. The high yield estimation in Virtuoso-ADE using the Worst-Case Distance (WCD) method is validated to estimate yield within 1% accuracy, with 200 parameter samples as compared to a traditional 10,000 random sample analysis. 

Assertions in Spectre/APS set in the Virtuoso-ADE environment are used to detect over-voltage conditions that can lead to device reliability problems. Circuit optimization capabilities in Virtuoso-ADE-XGL are validated for retargeting a generic ABS cell to a specific 28nm technology that meets given specifications. Additionally, a flow with Allegro SiP Architect and Virtuoso AMS-Designer was qualified for analysis of 3D-IC integration, leveraging ther Multi Technology Simulation (MTS) supported by Spectre.

Building on TSMC AMS v1.0 foundation in AMS v2.0, Cadence has delivered another strong set of capabilities for much more productive and predictable AMS design at 28nm. For further information on the TSMC AMS reference flow please contact me, your TSMC or Cadence representative.

Mladen Nizic