Workshop Minutes for BTI session 8:00 – 8:50 PM

Measurements

One point drop down is now industry standard, is there any need for any other method?

  • In some instances, a full “sense” Id – Vg sweep is not needed, and the single point measurement is sufficient.
  • Depending on where applicable (i.e., from a circuit perspective), ID,lin is important, which can dictate where the single point measurement is taken
  • New technologies/processes may require full sweep Id – Vg data to be able to observe all aspects of degradation
  • These new devices may require shorter sense delay times when incorporated with reduced overall stress time to minimize relaxation effects and provide the ability to observe early Vt instability during initial stress cycles.
  • Shorter overall stress time enables faster turnaround for screening new technologies/processes

What should be ideal range of stress bias (how close to Vdd), stress time and measurement delay?

  • For mature technologies, one can do more long-term stress time conditions that allow longer sense delay timesandnear Vdd stress.
  • Long-term stress times allow for stress bias values to be closer to use conditions thereby enabling more robust understanding of degradation in this voltage regime
  • Literature has demonstrated that sense delay time effects reduce significantly the longer the device is stressed
  • In addition, the first read out can be done at later times to also mitigate the effects of relaxation/sense delay time
  • For newer technologies such as PMOS with SiGe channel where the NBTI is lower, one has to apply relatively higher voltage to observe appreciable degradation.

HTOL tests often have long delay – concerns?

  • Yes, the potential long delay in HTOL would certainly be a concern.
  • There is some evidence that validates this concern
  • Temperature measurements “on the bench” was found to be more reliable
  • Ultimately, one needs to know make sure he/she knows what you need
  • Proper modeling is needed to include the delay impact

Is VT sense enough? Or need others (Idlin, etc)?

  • With current day measurement equipment one can extract multiple parameter (Id, gm, SS, Vt) in one go which can be useful for evaluating the newer technologies.

Modeling

What should a good model target?

  • A good model should consist of appropriate understanding (i.e., multiple stress bias values, temperatures, slew rate, work load, etc.) of how device is intended to be used.
  • Model should represent the outcomes of stress executed at typical workloads the device or circuit will experience
  • Depends on where devices are in the circuit for duty cycle dependence

Do we need TCAD to model device complexity?

  • Having a TCAD model would be considered useful.
  • If physics is built in, then model can be readily used to explain types of devices and/or device processing changes

SPICE models – are they adequate (provided by existing CAD vendors), what are the needs?

  • “Universal” SPICE models for BTI degradation would be nice to have.
  • Currently, organizations have their own PMIs since a universal model is not available
  • May be hard to have a universal model since, for example, each foundry has their own historical model that has been developed over multiple generations of their technology, thereby making it difficult to arrive at a standard/univerisal model

Any interest in including BTI in DTCO efforts?

  • BTI in DTCO, historically, has not been included.
  • If TCAD can properly implement BTI degradation, then it would be great to incorporate in DTCO
  • BTI incorporation could definitely be useful and relevant for novel devices because initial screening could to done to project if it will be a viable solution or not

Variability

Relative importance of time-zero and BTI variability?

  • The relative importance of time-zero and BTI variability depends on how big the time-zero variability is.

Are they completely uncorrelated?

  • Yes, agreed upon by the workshop attendees is that time-zero and BTI variability are uncorrelated
  • However, there is some evidence of a minor correlation.

Is RTN important relative to time-zero variability?

  • RTN is important, but depends on device geometry.

Does RTN change after stress?

  • Literature says RTN does not change after stress