print

How Reliable Is Your Full-Chip Reliability Verification?

One of the tasks during technology development is the development of a gate library that takes into account reliability issues like electrostatic discharge (ESD), latch-up (LUP), and time-dependent dielectric breakdown (TDDB) under as-designed operation voltage ranges. For interconnect technology, co-planar interconnect spacing requirements across dielectrics are determined primarily by optical and etching effects, with some consideration given to electrical parasitic impacts. Once the gate library and technology design parameters are developed and packaged into a process design kit (PDK), the place and route (P&R) tool uses them as input to P&R process, which is tuned against the customary cost functions to achieve design rule checking (DRC) compliance, timing, die size, IR drop, and electromigration (EM) design goals.

However, advanced processes enable many more devices to be packed into a die with equal or lesser dimensions than those of a previous node. For a 10 nm system-on-chip (SOC) design, it is not surprising to see multi-billions of devices contained in an area under 1 cm2. Based on the experience of foundries and integrated device manufacturers (IDMs), the P&R layout still requires full-chip reliability verification (and possible layout modifications) to prevent ESD along interconnect paths across multiple function blocks, LUP across gates in the vicinity of IO circuits, and TDDB between interconnects of different nets operated on different potentials. Such full-chip design verification practices are becoming critical for technology at and below 28 nm. Without it, chip reliability failures become far more likely during burn-in stage or in silicon.

FULL-CHIP RELIABILITY VERIFICATION
During full-chip design, cells and transistors are connected through interconnect metal. Because the physical circuitry is extremely condensed in advanced designs, and it often operates in multiple power domains, new reliability concerns have emerged for the interconnect at both the chip and block levels.

From a strict technology perspective, the foundry or IDM always provides a reliability design kit to circuit designers. However, in advanced node foundry/IDM technology kits, some critical reliability design rules are often defined simply as guidelines, with little consideration given to the feasibility of implementing them in an EDA tool. To meet stringent reliability expectations while maintaining time-to-market schedules, designers need efficient and accurate automated full-chip reliability verification techniques and tools.

DYNAMIC SIMULATION
Dynamic simulation is a technique often used in various reliability verification approaches, mostly at the cell/transistor level, to detect and prevent reliability issues. Designers frequently choose the familiar dynamic simulation approach for full-chip reliability verification, only to find the task is impractical. Using dynamic simulation approaches to analyze ESD, LUP, TDDB, and electrical overstress (EOS) problems is not feasible for block or whole chip designs, primarily because the large scale of the circuits produces too much data to be processed to provide reliable simulation results in a practical runtime.

STATIC DESIGN RULES
In an attempt to facilitate automation and still meet rule requirements, the use of static design rules is becoming a realistic and effective approach to manage reliability issues in the chip/block-level design stage. For example, to prevent weak interconnect spots being burned by high ESD current surges, the typical foundry/IDM rule defines the required wire width of the ESD path to be larger than some criteria to survive an ESD event. Based on this guideline, fabless layout designers usually try to use a traditional DRC tool to check the wire “width” against the rule criteria. However, a traditional DRC tool can’t identify the direction of the electrical current path along the polygon shape (which is needed to determine the “width” of the path), and it doesn’t know how to process the individual weighted widths of a parallel wires network to check against rule criteria. To support their customers’ needs, foundries/IDMs are responding with more clearly-defined static current density checks along ESD paths to enable designers to find these weak interconnect spots.

To prevent LUP issues, the foundry/IDM historically provided a DRC approach that performed both a spacing check and a check for the presence of a guard ring between aggressor and victim circuits at the chip assembly design stage. However, to enable a DRC tool to recognize the polygons that are a part of these checks, designers often manually placed markers on designated polygons. In an iterative design process, it is difficult for designers to find a reliable and consistent way to place markers at the appropriate locations on the layout. The marker methodology requires designers to identify the circuits behind ESD resistors (connected to IO Pads) that are involved in a LUP check. At the same time, the LUP sensitivity of these circuits depends on the effective resistance of the ESD resistors. If the effective resistance (a value the DRC tool doesn’t and can’t know) is bigger than the rule criteria, then circuits behind the ESD resistors are considered to be isolated from LUP. Some markers also require the identification of delta-operation voltage on polygons of different nets. However, depending on the delta- operation voltage range, the spacing check criteria will differ. Foundries and IDMs are now defining non-physical marker methodology that allows logic-driven layout (LDL) checking functionality to find LUP circuits/nets that can be further transformed to physical shapes for a property-annotated DRC check.

When verifying interconnect TDDB, the foundry/IDM provides a DRC approach to check the spacing of polygons on different nets at the chip assembly design stage, where the spacing criteria is dependent on delta-operation voltage. However, this methodology also uses multiple physical markers representing different voltage values. Designers struggle to place markers of the appropriate voltage values on polygons in nets of concern across the whole layout. Not only do the physical markers not work well with delta-operation voltage dependent spacing criteria, but assigning voltage from IO/power/ground pads to nets associated with inner circuits is difficult. Instead, foundries and IDMs are now defining voltage text/annotation methodology that allows LDL checking functionality to propagate the voltage values into the entire circuit design, and export nets of TDDB concern, for which the logical nets can be further transformed to physical shapes for property-annotated DRC checks.

Applying static design rule checking to earlier design stages for block or intellectual property (IP) reliability verification can also help reduce design iterations. While the static approach may yield a small amount of over-pessimistic values (especially for TDDB), its ability to detect complex reliability issues early in the design flow outweighs the reasonable amount of manual work needed to evaluate these values prior to tapeout.

LOGIC-DRIVEN LAYOUT CHECKING
Logic-driven layout checking has the unique ability to use both netlist and layout (GDS) information simultaneously to perform electrical checks that incorporate both layout-related parameters and circuitry-dependent checks, enabling designers to address these complex verification requirements. Using static simulation and static voltage propagation in conjunction with an LDL flow, foundries/IDMs can now clearly define ESD and LUP rules with assurance of 100% coverage based on their rule requirements, and design companies can implement automated full-chip ESD, LUP, and TDDB rule checking. The LDL flow for both SPEF and SPICE networks is shown in Figure 1. This flow replaces the manual marker flow and provides automated reliability verification.

Figure 1: Steps in an LDL flow for both SPEF and SPICE networks.

AUTOMATED FULL-CHIP RELIABILITY VERIFICATION
For ESD prevention, diode/MOS/resistor as ESD or power-clamping device clusters with enough strength must be connected to IO, power, ground, and cross-power domain paths. Automated reliability verification provides the functionality to traverse connectivity inside a SPICE netlist to verify that the required (foundry or user-defined) ESD/power-clamping/back-to-back diodes circuits exist or not. Device parameters of all corresponding circuits are also examined to ensure the devices (individual/clustered) have sufficient strength for adequate ESD protection.

For advanced technology, the influence of interconnect parasitics is becoming significant. Of concern is their impact on ESD across the whole chip, not just whether or not the ESD protection circuits exist. To ensure ESD paths (connected by interconnect metals) function as designed, it is critical to examine not only the placement of an ESD circuit, but also the effective resistance along the ESD path. ESD paths are defined by either the foundry or the designer, and typically contain multiple P-2-P segments. The “P” can be either a device pin or a cell port. Each P-2-P segment is formed by pin pairs on same net, and pin pairs can be exported with an annotation index for grouping purposes during simulation. The pin pairs will be transformed into probe points in SPEF network. Automated reliability verification provides functionalities to search ESD paths inside a SPICE netlist (Figure 2), extract non-redundant and reduced SPEF networks for corresponding ESD paths, then run static simulation (in a user-defined grouping mechanism) to calculate effective resistance of ESD paths meeting criteria (foundry or user-defined) or not.

Figure 2: Searching ESD paths in a SPICE netlist.

Even when the electrical effective resistance meets criteria (usually an amount less than some value in ohm), a huge ESD surge current (with different dynamic natures for different ESD modes) can easily burn a weak interconnect spot. To prevent this, the current density along ESD paths through all wire segments and via arrays should be examined to ensure current density is less than some tolerant value. Designers can use automated reliability verification functionality to calculate DC current densities along ESD paths. The flow is similar to determining the effective resistance of ESD paths described above, but the SPEF network is un-reduced to preserve physical polygon data for each wire segment and via area to enable calculation of the current density flowing through them.

In addition to inserting guard-rings/guard-straps, LUP prevention requires that spacing among polygons of active diffusion, P+ active, and N-well device layers operated at different potentials should be equal or larger than a criteria for which the value increases with increases of potential difference (delta- operation voltage). This delta-voltage-dependent spacing requirement for LUP checks is considered to increase the reliability of a design. The circuits/polygons scheme for LUP checking is shown in Figure 3. The spacing criteria and delta-voltages are usually grouped in a few bins, with one spacing criteria assigned to one delta-voltage range. A user input file is needed to assign the operational voltage value to corresponding IO/power/ground ports.

Figure 3. Inside a LUP sensitive zone, safe spacing between active diffusion, P+ active, and N-well polygons is dependent on delta-voltage (Va-Vb and Va-Vc) range.

An automated LUP checking flow performs the following steps:

  1. Extract the layout netlist (standard circuit verification step provided by foundry/IDM).
  2. Traverse connectivity graph, propagate voltage values from IO/power/ground ports into internal nets based on user-defined constraints, and locate and export potential aggressor/ victim devices annotated with a propagated voltage value. Voltage propagation is executed without alternating the voltage value. This voltage propagation behavior is reasonable for a LUP check, since the potential LUP issue occurs close to an IO circuits region.
  3. Generate unique physical layers that correspond to exported aggressor/victim devices. These layers are usually a collection of polygons overlapped with device formation seed shapes. This procedure also annotates the voltage value as a polygon property.
  4. Process polygon data with annotated properties, and perform spacing DRC checks to find any edges or polygon pairs that violate spacing criteria according to the delta-voltage range.

For interconnect TDDB checks on block or full-chip designs, spacing checks among polygons of the same interconnect layer on different nets are executed against criteria dependent on delta-voltage range. The challenge is to propagate voltage value from IO/power/ground ports to internal nets, such that all targeted nets for potential TDDB concerns have the appropriate voltage value to annotate to their corresponding polygons. Unlike voltage propagation for LUP checks, the voltage values starting propagation from IO/power/ground ports shift magnitude when going out from some level-shifter-related circuits. Unless these voltage shifts are handled appropriately, users will eventually see huge quantities of overly-pessimistic values. The static voltage propagation engine doesn’t usually recognize this level of circuit function automatically. It requires the user to design a static voltage propagation scheme that includes additional information—like voltage values on some cell ports related to level-shifter circuits, or sub-circuit pattern definitions—to enable the voltage propagation engine to manipulate voltage shifts.

An automated reliability verification platform should contain functionality to perform static voltage propagation under a user-constrained propagation scheme, assign appropriate voltage values to each concerned net, annotate static voltage values to physical polygons of corresponding nets, and run spacing checks that account for different spacing criteria versus different delta-voltage ranges. An example of a static voltage propagation scheme is shown in Figure 4. The top port has a 3.3 voltage. Using static voltage propagation, 2.5, 1.8, or 1.2 V can be assigned to other nets inside this circuit, depending on the design of the voltage propagation scheme. Voltage shift is executed by user-defined sub-circuit patterns, or by acquiring the voltage value on the output cell port of a level-shifter circuit by means of simulation. After completing static voltage propagation, the LDL flow is used to annotate the voltage values to polygons in nets of concern for DRC checking.

Figure 4. Automated static voltage propagation.

There are multiple means by which designers can acquire cell port voltage data. For example, a simplified level of dynamic simulation can be executed to acquire such data. A static voltage propagation engine can be instructed to propagate voltage, break out when encountering the above cell ports or sub-circuit patterns, and then continue propagation based on acquired voltage data on cell port or voltage constraints defined with the sub-circuit patterns.

In a real-world application, if the voltage shift scheme/function is well-defined, designers will see relatively clean run results. There may be some small quantity of overly-pessimistic voltage annotation values, due to the nature of static propagation. However, the number of these values is typically in a human-manageable range, so they can be manually processed toward final tape-out without any significant effect on schedules.

SUMMARY
Advanced nodes are introducing new and complex reliability conditions that can’t be easily or accurately checked using dynamic simulation or traditional physical and circuit verification technology at the full-chip level. A new approach that employs static simulation and static voltage propagation in conjunction with logic-driven layout analysis supports the development of accurate, fast, automated reliability design verification for ESD, LUP, and TDDB issues. With new reliability verification methodologies and tools, designers can now verify that their designs are protected against a wide range of reliability issues, ensuring that the final product provides the performance and product life the market demands.

 

 

 

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • TwitThis
Extension Media websites place cookies on your device to give you the best user experience. By using our websites, you agree to placement of these cookies and to our Privacy Policy. Please click here to accept.