Multi-gb/s ic test challenges and solutions

Abstract

Production test and device I/O characterization are different tasks. Measurements and analysis associated with these different tasks need quite different tools. 1 Serial Data Links -the future is “bright” The 2001 ITRS predicted serial datacom rates reaching 40 GBPS by this year, doubling within a decade - and 40 GBPS production devices now exist. A 12.5 ps unit interval is tiny by today’s standards; for reliable data transmission to reach these rates the challenges are daunting. Despite this, we can fully expect such serial data links. What is required for them to be producible? Quite simply: reliable design, consistent process technology, and solid production test capability. 2 Design Qualification vs Production Test It is absolutely necessary to distinguish clearly between design qualification/characterization and the sort of test necessary to guarantee that a particular device is manufactured properly. Yet to sort one from the other is no mean task. It only takes a few extra ohms of interconnect to marginalize a good design; these few extra ohms will displace a transition by a few but unacceptably large number of picoseconds. Each bit in a high-speed serial data stream can be characterized by two key parameters: unit interval and individual edge displacement, expressed as a fraction of unit interval. Edge displacement is often referred to as jitter. Jitter is an unfortunate term, as it masks what is actually going on. Yes, some edges are out of place, but why are they? Displaced edges are created by intersymbol interference, internal systematic path or propagation delay errors, clock phase error, and random noise. These errors must be identified and minimized during the design qualification phase; the design is acceptable if error rate is sufficiently below target unless the product is defective. Key point: “random noise” in semiconductors is not very large. RJ in excess of 1 ps is more likely to be instrumentation noise than true circuit noise, or else is uncorrelated systematic error. This error is likely to be correctable in design. Laboratory equipment is generally capable of acquiring the data necessary to characterize today’s I/O designs. However, it is a costly mistake to attempt to translate these characterization tasks into a production test requirement. Measuring “jitter” in a production test environment is unneeded. What might be measured (and compared to design minimum, nominal, and maximum) are bit-by-bit displacements focused specifically on the structure of the specific serializers or deserializers in the signal path. This requirement is derived from the concept of defect-based testing: each element in the circuit (whether transistor, passive component, or interconnect) has a target quality requirement; if it is not manufactured to this quality requirement then something - something specific - is wrong. Yet this overall approach is insufficient by itself. It misses the tight coupling of the serial data stream to its internal self-aligned clock. If this clock is too erratic, then perfect serializing or deserializing circuitry will break. Conclusion: production test has two requirements: qualify the clock and qualify the bit generators or extractors. There should be no need for long data streams, or concomitant long test times.