A few years ago, I led the development of a relatively small IP in the SERDES application space. The design effort was measured in weeks, with about three months applied to verification. We then engaged our first customer who wanted to replace the transmitter side of an existing solution provided by a market leader. We were confident on our verification efforts, and helped the customer deploy our IP using an FPGA to demonstrate an end-to-end working prototype. Then trouble started. Two months later, countless hours of debugging and lengthy three-way engineering meetings, we finally discovered that the third-party receiver was not 100 percent compliant with the standard. A 5 minutes change to the IP resolved the issue, marking the end of major engineering effort of discovery, and a management nightmare.
"When dealing with IP, you are mainly buying a head-start on verification"
Sometime earlier I had been involved as a customer in the acquisition of a silicon-proven IP with an outstanding track record and close to a billion parts in the market. The expectation was set for minimal verification. Months later, and countless hours of debugging by the silicon design team, the OS team, and Application Software teams, we were able to determine that the IP had a bug that prevented us from releasing our product. What apparently passed as an unnoticed failure for the existing hundreds of millions of parts using the IP in other products, turned out to be a showstopper issue for us, and once again a major engineering effort of discovery.
Plenty of similar examples exist in the past 25 years, where I have been both, at the receiving and delivering ends of the Semiconductor IP business, dealing with well over one hundred IPs. My experience can be summarized as “When dealing with IP, you are mainly buying a head-start on verification. The cost of the IP does not necessarily correlate with how far ahead you are actually getting.”
IP complexity and the need for differentiation in the marketplace have been main drivers undermining success with IP. On one hand, feature-rich demand and complex standards have expanded the verification challenge to the point that full coverage is only possible for the simplest of IPs.
Differentiation has brought an increase in customization and, in some cases, deviations from standards, creating problems for both, IP customers and providers, which in most cases need to significantly augment their verification efforts. It is at this point where the cost/benefit trade-off between acquiring versus developing IP comes into question.
As I look back at my long past experience in this space, I realize that my number one criteria for selecting IP has been the expertise of the provider, and the single most important learning: “buy support”.