Energy optimal sizing of FinFET standard cells operating in multiple voltage regimes using adaptive independent gate control

Yue Fu, Yanzhi Wang, Xue Lin, Shahin Nazarian, Massoud PedramACM Great Lakes Symposium on VLSI 2014 · DOI

TL;DR

This paper introduces a novel design framework for FinFET standard cells that optimizes energy efficiency across a wide range of operating voltages, from subthreshold to superthreshold. By adaptively controlling the independent gates of FinFETs, the authors achieve significant reductions in energy-delay product, making ultra-low power designs more practical.

What Problem Does This Paper Address?

As semiconductor technology scales down, traditional CMOS devices face limitations in power consumption and performance. FinFETs, with their superior gate control, offer a promising alternative, especially for ultra-low power applications. However, FinFETs exhibit significantly different characteristics in subthreshold (very low voltage) and superthreshold (normal operating voltage) regions. This disparity makes it challenging to design standard cells that perform optimally across all operating regimes. Specifically, achieving a balance between energy consumption and circuit delay, while also ensuring balanced rise and fall times, becomes difficult when operating at very low voltages where FinFETs are most power-efficient but also slowest. This paper addresses the need for a unified design approach that can leverage the unique properties of FinFETs to achieve energy-optimal performance across this entire spectrum of operating voltages.

Methods

The core of this paper's contribution lies in its proposed design framework that utilizes an "adaptive independent gate control" method for FinFET standard cells. This method exploits the unique feature of FinFETs where the front and back gates can be controlled independently.

  1. Transregional FinFET Model: The authors first employ an analytical transregional FinFET model that accurately captures the device behavior in both subthreshold and near-threshold regions, as well as the superthreshold region. This model is crucial for accurate simulations across different voltage regimes.
  2. Adaptive Independent Gate Control: The key innovation is the adaptive control of the independent gates (specifically, the back gate voltage, VBN). Instead of a fixed back gate voltage, this approach dynamically adjusts VBN based on the supply voltage (VDD) and the desired optimization objective.
  3. Optimization Objectives: Two primary optimization objectives are defined:
  4. Achieving equal rise and fall times (which minimizes circuit delay).
  5. Minimizing the energy-delay product (EDP), a critical metric for power-constrained designs.
  6. Design Framework: A robust FinFET standard cell library is designed using this adaptive control method. The framework derives optimal back gate voltages for different supply voltages to meet either of the optimization objectives.

Datasets

The paper evaluates its proposed framework using:

  • ISCAS Benchmarks: Specifically, the C432 and C499 benchmark circuits are used to demonstrate the effectiveness of the adaptive independent gate control on more complex logic.
  • 32nm Predictive Technology Model (PTM) for FinFETs: This is a standard technology model used for simulations, providing realistic device characteristics for the target technology node.
  • Inverter Chains: A 20-stage inverter chain is used as a fundamental building block to illustrate the delay and energy-delay product optimization.

The paper does not introduce new benchmark circuits but rather applies its methodology to existing, well-established ones.

Key Insights and Ideas

  • Subthreshold is King for Ultra-Low Power: The minimum energy point for FinFETs lies in the subthreshold or near-threshold region, significantly lower than for bulk CMOS. This is where the most substantial energy savings can be realized.
  • Independent Gate Control is a Game Changer: The ability to independently bias the front and back gates of a FinFET provides a powerful knob for trading off delay and energy. This is particularly effective at lower supply voltages where device characteristics are less predictable.
  • Adaptivity is Crucial: A static back gate bias is insufficient. The optimal back gate voltage needs to adapt to the supply voltage (VDD) to achieve consistent performance across different operating regimes.
  • Balancing Act: Optimizing for minimum delay (equal rise/fall times) doesn't automatically yield the minimum energy-delay product, and vice-versa. The adaptive control allows designers to target either objective.
  • Activity Factor Matters: The benefits of adaptive independent gate control are more pronounced when the circuit activity factor (how often signals change) is higher.

Key Results

The proposed adaptive independent gate control framework demonstrates significant improvements:

  • Delay Reduction: Up to 48.5% reduction in circuit delay for a 20-stage inverter chain in sub/near-threshold regions compared to circuits without independent gate control.
  • Energy-Delay Product (EDP) Reduction: Up to 64% reduction in EDP for benchmark circuits when using the proposed optimization framework.
  • Voltage Regime Benefits: The most significant reductions in both delay and EDP are observed at lower supply voltages (subthreshold and near-threshold regions), highlighting the framework's effectiveness for ultra-low power applications.
  • Model Accuracy: The empirical model used for FinFETs in sub/near-threshold regions shows a maximum inaccuracy of 7.76% compared to HSpice simulations.

Limitations and Future Work

  • Model Simplification: While the transregional model is accurate, it's an empirical model. Further work could explore more physics-based models for even higher fidelity.
  • Scope of Benchmarks: The evaluation is performed on a limited set of ISCAS benchmarks and inverter chains. Testing on a wider variety of complex circuits and application-specific designs would provide broader validation.
  • Process Variations: The current work assumes ideal device characteristics. Investigating the impact of process variations on the effectiveness of adaptive independent gate control would be a valuable next step.
  • Dynamic Adaptation: The current framework focuses on adapting VBN based on VDD. Future work could explore more dynamic adaptation strategies that respond to real-time circuit activity or temperature variations for even greater optimization.

Summary updated

BibTeX
@inproceedings{fu2014energy,
  author = {Yue Fu and Yanzhi Wang and Xue Lin and Shahin Nazarian and Massoud Pedram},
  title = {{Energy optimal sizing of FinFET standard cells operating in multiple voltage regimes using adaptive independent gate control}},
  booktitle = {ACM Great Lakes Symposium on VLSI},
  year = {2014},
  doi = {10.1145/2591513.2591555},
}
Figure