Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Figure 1: Leakage power increases with advanced technology Given the growing concerns around managing power consumption, design teams are employing advanced low-power design techniques to minimize power and help with the cooling issues
associated with large designs. Advanced low-power design techniques include the use of multiswitching threshold transistors (Multi-Vt), using multiple power supply domains with multiple voltages (MSMV), power shut-off (PSO) and dynamic voltage and frequency scaling (DVFS). Often, a combination of these techniques must be used to meet the required power targets.
Multi-Vt Transistors
One of way to manage leakage involves using transistors with different switching thresholds. Low-threshold transistors enable high performance but suffer from high leakage. On the other hand, high-threshold transistors suffer less leakage but offer less performance. Advanced standard cell libraries now contain cells that are optimized for power or optimized for timing. These cells enable design implementation solutions to automatically optimize timing performance, while minimizing leakage power.
Thermal Analysis
On-chip temperature control has been a known problem for some time; this mainly involves using special cooling techniques associated with the packagesfins, fans, liquid cooling, etc. In general, packages with higher cooling characteristics cost more, so minimizing on-chip power can result in significant cost reductions if a cheaper package can be utilized.
Focus around on-chip temperature variations is somewhat new, and increasing with the use of multiple modes of operation. Thermal variations directly impact timing and SI analysis by changing device characteristics, thereby causing additional setup/hold time violations. On-chip thermal analysis also enables a design team to understand if high leakage components are located in high temperature areas; relocation of such components could result in a cooler chip that requires less expensive packaging.
Manufacturing Effects
On top of all the complexities associated with creating the design itself, design teams must also account for the complexities associated with the manufacturing process, because small variations in the process can result in large variations in timing, power, and signal integrity. Multi-mode, multi-corner (MMMC) analysis is becoming more common, where chip performance is validated for many modes of operation at multiple process corners. As the number of corners increases, a statistical approach is necessary to bound the runtimes. Statistical static timing analysis (SSTA) is already in use for advanced designs, and there is a growing focus on statistical leakage power analysis (SLPA) to better optimize power across all process corners. The end result is that most aspects of designing and verifying modern IC designs are becoming increasingly challenging, while the cost of missing the schedule continue to grow. Data from the Global Semiconductor Association (GSA) shows that 89% of designs miss their deadlines by an average of 44%, and that the cost of being even three months late can equate to as much as $500M in lost profits.
network). At the early floorplanning stages, comprehensive design data is not typically available, so the power analysis solution must provide the best accuracy with the available data. During physical implementation, design teams are trying to complete the placement of the blocks/cells and optimize timing. At this stage, most of the design data is available, and so a more accurate analysis can be accomplished. Because achieving timing closure is tough, a fixed operating voltage is often assumed at this point the challenge is to create power rails that meet these requirements. Once a design is getting close to tapeout, comprehensive, full-chip, full-accuracy analysis is required to try to catch any remaining issues that might cause silicon failure or yield issues. There are five major requirements a power/timing/signal integrity analysis solution must support; the following section discusses these requirements in more detail.
ECO
Power Switches
Power Switches
VDD
Power Switches
VDD
Power Switches
Figure 3: Power switches connect the block to the un-switched VDD power rail. When the chips Power Management Unit (PMU) detects that a block is not currently needed, it initiates a series of events to power-down the block. Before any power-down can occur, however, surrounding logic must be isolated and/or stabilized to prevent possible data
corruption. Once the surrounding logic has been isolated and/or stabilized, the block is powered-down by deactivating the power switches. Once the PMU logic detects that a block is again required, it triggers another sequence of events that start the powering-up cycle. Powering-up a block can demand a large, dynamic current, known as rush current, and design teams need to make sure that this rush current does not create problems in the surrounding logic by causing local IR drop. Controlling the amount of rush current is achieved by changing the number and size of the power switches and the sequential delay between them, as shown in figure 4.
Rush Current
Time
Rush Current
Time
Figure 4: High rush current (top) versus lower rush current (bottom). An advanced power integrity solution helps a design team determine how to optimize the power switches. Too many power switches can increase rush current, while too few can increase the IR drop seen by the logic. There needs to be a careful balance. The planning and optimization of power switches should be driven by detailed IR drop analyses, where steady state analysis shows the IR drop when the logic is operating normally, and dynamic power-up analysis shows the impact on surrounding logic when blocks are re-activated. In practice, the number/size of the power switches is determined by the maximum tolerable IR drop, and the sequencing delay is used to manage the rush currents.
Figure 5: Clock Jitter due to IR drop-induced delay variation. Clock Skew analysis identifies the difference in arrival times of the launch and capture registers as illustrated in Figure 6. Clock Skew analysis uses high activity IR drop for late path delay calculation and low activity IR drop for early path delay calculation.
Figure 6: Clock Skew due to IR drop-induced delay variation. In addition to IR drop induced clock skew, IR drop on the gates within the signal paths causes delay variation for the signals, which in turn causes additional setup and hold time violations. Hopefully, it is now clear why it is increasingly important that IR drop-aware static timing analysis be performed, to account for the impact of IR drop on clock jitter, clock skew, and setup and hold times.
Summary
Design teams are increasingly facing parallel challenges of minimizing power while at the same time continuing to close timing and avoid signal integrity issues under complex conditions. While timing closure remains the number number-one issue that challenges most design teams, power management has rapidly ascended to become the number number-two concern. Physics dictates that power, timing and signal integrity are all inter-related, and that independent analysis of each of these components will not catch all of the potential issues. The conclusion is that effective management of timing, power, and signal integrity requires an integrated analysis solution within the design environment, enabling analysis from early floorplanning to pre-tapeout signoff.
Pete McCrorie is currently focused on product marketing for power rail analysis and mixed signal implementation at Cadence. Before joining Cadence, McCrorie was managing all extraction and analysis products at Simplex. McCrorie has his Masters degree in Physics with Electronics from Liverpool University, UK. Harish Kriplani is currently an R&D Group Director at Cadence, where he is responsible for architecting and managing solutions for power/power grid analysis and design. Kriplani received his BTech from IIT Kanpur and PhD from University of Illinois at Urbana-Champaign and holds 5 US patents.