The Machine Model test, as a requirement for component ESD qualification, is being rapidly discontinued. Here we illustrate why MM evaluation is not necessary for qualification. The following major conclusions can be made about MM in general:
Background. The Machine Model (MM) is a widely misunderstood component ESD qualification test method. It continues to generate confusion for OEM customers and their IC suppliers during ESD qualification. Many companies and design organizations continue to use MM, mostly as a legacy “required” practice, despite the fact that it has been downgraded by three standards bodies and is no longer recommended for qualification testing in accordance with Jedec JESD47. The automotive industry, a longtime user of this method, no longer requires it in their AEC-Q100 list of qualification tests. The scopes of the Jedec (JESD22-A115) and ESDA (ANSI/ESD STM5.2) test method documents have also been changed to reflect this status. There are a number of reasons for these changes, as will be outlined below. The continued use of MM for qualification based solely on legacy requirements has no technical merit, given the information that has been gathered over the past few years. Those companies that continue to use MM will take on an unnecessary and burdensome business approach. The reasons against use of the MM are below:
MM vs. HBM and CDM. The waveforms for HBM, MM and CDM are compared in Figure 1. The HBM and MM have similar ranges of rise time (2 to 10ns). Therefore, any thermal heating in silicon taking place in this time period leads to the same failure mechanisms for both models. This holds true for all technologies, including advanced technology nodes. This early part of the waveform determines where protection circuits must be deployed in design. With similar rise time characteristics, HBM and MM encourage the same protection designs. For CDM, on the other hand, the rise time is much faster (0.1 to 0.5ns) and often leads to a unique failure mechanism like oxide breakdown. Even more important, the observed ESD field failures are dominated by oxide breakdown when the CDM level is not adequate. Thus, a different set of protection strategies are generally needed for CDM. This makes it more critical to focus on CDM qualification, instead of duplicating the HBM test information by using the MM. Figure 1 shows the observed failure modes for the same I/O pin after stressing with HBM, MM and CDM. It is clear that with HBM and MM, the damage sites were the same, in the protection diode, but with CDM stress, the damage site corresponds to oxide breakdown in the output transistor. This also illustrates that meeting high levels of MM does not improve CDM performance until the right effective design techniques are employed.
Commercial MM testers have inductors built into the MM stimulus circuit. These inductors must be present to produce the oscillatory waveform required in the MM test method. The inductors, however, actually slow the MM waveform (Figure 1), and therefore, MM cannot represent very fast metal-to-metal contact discharge as CDM does. On the other hand, the CDM test is directly represented by elevating the package potential and directly grounding the pin to produce the fast discharge. MM cannot be relied on to accurately model fast metal-to-metal contact discharges, which are known to occur in the field.
Metal discharge vs. CDM discharge. Analysis of Tanaka4 is shown here to demonstrate that a metal discharge from a small metallic object to a device is similar to the commonly used CDM test. Tanaka considers small objects because large machines (typically >10pF) are almost always grounded for reasons beyond ESD and thus pose little practical threat for these events. On the other hand, tools and small machines are difficult to ground and may lead to charging effects, where the capacitance of the metal object is related to surface area and distance. These values can range from <1pF to nearly 10pF. For example, this could be as much as 1pF for a small metal object of 10cm2 at a distance of 0.5cm. Both the small metal discharge and the CDM discharge can be represented by the same set of equations for I(t), and thus both can be expected to generate the same discharge event if the values of the parameters are similar. Figure 2 illustrates the case for a small object of 10pF for both metal discharge and CDM discharge.
The above analysis is confirmed by measurements3 (Figure 3) where the discharge (a) from a charged tweezer to IC pin is the same as direct discharge from metal as shown in (b), and both are similar to the generated CDM discharge in (C). The time scale for both metal discharge and CDM discharge are indeed the same, clearly indicating that CDM is a good representation of the metal discharge in the EPA.
The Industry Council on ESD Target Levels has studied the HBM and MM results on a wide variety of designs in many technologies and has concluded that MM is intrinsically related to HBM, with a correlation factor “range” that is dependent on the HBM design level.2 These data are represented in Figure 4. However, the most important conclusion of the study was that MM is a redundant test and that a sufficient level of MM robustness is automatically included in an adequate HBM design. This also includes the bipolar nature of the MM stress. Any oscillatory waveform that might be measured during discharges in the field is sufficiently covered, if the part is proven to have an adequate HBM design.
This minimum design value, as measured by an MM tester, is well above any voltage remaining on all properly grounded machines in an ESD protected manufacturing environment. In essence, meeting a safe value for HBM (and CDM) is sufficient for production of ICs without needing to evaluate MM as an additional qualification.
Field data analysis. The Industry Council work has shown that most of the overstress field returns exhibit failure signatures of higher energy EOS, and that the level of HBM ESD from 500V to 2000V (shown as the HBM Failure Analysis Return (FAR) window in Figure 2) on 21 billion shipped units did not show a correlation to the customer field return rates. Similarly, these very same shipped units (500V to 2kV HBM) also had MM levels in a range of 50V to 300V, as also seen from Figure 2. Therefore, it can be concluded that the EOS field returns are indeed not related to this range of intrinsic MM levels. That is, it does not matter if a shipped device has a measured MM value of 50V or 300V.
Devices with various measured MM levels have shown no correlation to real world EOS failure returns.
Standards bodies and positions on MM. During the past two decades, the electronics industry’s standards bodies have changed their viewpoint with regard to MM and its requirement for IC qualification. At present, JEITA does not recommend MM. The Automotive Electronics Council’s AEC Q100 standard gives a choice between HBM and MM, but does require CDM. In recent years, Jedec has strongly recommended discontinuing use of the MM for ESD qualification because of its test variability and non-correlation to real world failure modes. In general, standards bodies have come to recognize that:
The following statements are from the Jedec website:
Conclusions
The information herein supports the discontinuation of MM as part of IC qualification. Most important is that a wide range of products having only HBM and CDM testing performed are being shipped today at volume levels in the billions, with no field returns due to ESD. These products, passing at or above the recommended minimum HBM and CDM levels, are being routinely shipped by major suppliers, and are accepted by major OEMs. No increase in field return rates has been observed with MM removed from qualification for these products.
The confusion generated by MM has persisted for over two decades. The presumed need for this test is causing additional qualification delay due to an extraordinary consumption of design / test resources, added delays in time-to-market, and in some cases having an impact on IC speed and performance. Maintaining safe HBM and CDM levels is sufficient to meet all IC manufacturing, handling and assembly needs.
Different customer sectors may feel they need enhanced ESD requirements for specific reasons. For example, some automotive customers have more consistently required MM model testing, the impression being that an independent and redundant test provides enhanced safety, improved quality and reduced defectivity. However, industry experience has shown that passing a redundant (to HBM) MM qualification test does not help automotive manufacturers achieve these goals. Meeting current industry standard HBM/CDM will ensure that a product can be safely handled with sufficient margin to prevent ESD damage and maintain the quality/reliability of the product as shipped from the component manufacturer. Since many suspected ESD failures turn out to be higher energy EOS in nature, methods to prevent electrical overstress during manufacturing will also help maintain product reliability.
Common goals. We have presented evidence and arguments that the MM test of ICs is redundant, and there is no proof that devices have failed in the field because MM evaluation was not done. We strongly recommend that this test be discontinued for ESD qualification. This will save the semiconductor industry a tremendous and unnecessary burden by greatly reducing the routine characterization that is done to support the qualification process. The ESD robustness designed into integrated circuits to survive HBM and CDM testing will provide protection against any MM-like stress. Eliminating MM testing of ICs has no deleterious effects and will free up resources for more important engineering challenges.
References
1. M. Tanaka, JEITA/JEDEC Meetings, September 2011.
2. JEP155, “Recommended Target Levels for HBM/MM Qualification,” jedec.org and esdtargets.blogspot.com.
3. M. Tanaka, K. Okada and M. Sakimoto, “Clarification of Ultra-high-speed Electrostatic Discharge and Unification of Discharge Model,” EOS/ESD Symposium, 1994.
4. ESDA standards document definitions and hierarchy are summarized at esda.org/documents.html.
Ed.: This article is courtesy of the ESD Association (esda.org).
Charvaka Duvvury, Texas Instruments (ti.com); Robert Ashton, ON Semiconductor (onsemiconductor.com); Alan Righter, Analog Devices (analog.com); David Eppes, AMD (amd.com); Harald Gossner, Intel (intel.com); Terry Welsher, Dangelmayer Associates; and Masaki Tanaka, Renesas Electronics (renesas.com).