The recurring semiconductor industry cycle between standardization and customization was accurately described by Tsugio Makimoto in his presentation entitled "Implications of Makimoto's Wave" in the early 90s. From the mid-90s until approximately 2010, Makimoto's Wave has been off its normal frequency -- stalled in the standardization phase.
Today, the cycle has shifted dramatically toward customization. Apple's development of its custom A4 application processor used in the iPad and iPhone signaled the beginning of this cyclical shift and placed system companies, rather than semiconductor companies, squarely in the drivers' seat for development of complex, multicore systems-on-chip (SoCs).
This shift in the semiconductor cycle and control over SoC designs is due in part to the cost of overdesigning chips to serve multiple applications and part to failing economic models. For example, in the cell phone and digital TV markets, semiconductor companies attempted to build increasingly larger SoC platforms targeting an ever broader set of applications in order to generate the consumer volumes necessary to turn a profit. The economies of scale promised by standardized semiconductor platforms fell apart as the designs became too monolithic and the cost for a single development program reached upwards of $200 million.
During the peak of the last standardization cycle, it was not unusual for a semiconductor company program to employ a dedicated hardware design team of 1,200 people with an even larger software development group. The two things that broke the back of that approach were:
- Investing $200 million to sell a standard device for $10/unit means you have to sell many millions of chips before you ever recoup the upfront NRE. Inevitably, that model limited the number of successful suppliers even with a potential billion unit cell phone market.
- The rise of specialization exemplified by the SoC that Apple designed for the iPhone changed the economics of the cell phone application processor business. Previous to Apple, the area of an application processor was based upon a static price model that supported designs of about 50 square millimeters. Apple believed it could provide a more compelling user experience by substantially improving the performance of the graphics sub-system even though the resulting SoC used 2.5 times more silicon area than traditional application processors. The rest is history.
The emerging markets of IoT and wearables also exhibit this trend toward customization and specialization. Here, it is not the economics driving higher levels of SoC integration, but more the packaging and power requirements for these systems.
Traditionally, the purpose of SoC integration was to reduce cost or improve performance. But, IoT and wearables designers must develop more highly integrated solutions in order to enable the actual application. They can't demonstrate the value of their products if they don't first get to a low enough power level so the chip fits into a small enough form factor that consumers are willing to wear, use, and provide the critical feedback necessary for the system company to improve the next-generation device. Designers need highly integrated silicon because they can't prove the viability of the end system device without it. In this customization cycle, it is the systems companies with direct access to the end customer that are in the best position to mine the data and take the risk to develop SoCs for wearables.
In the context of a $700 smartphone or a $1,000 TV chassis, the system company has more room to make choices between a $10, $20, or $30 chip in the BOM -- optimizing their system profit instead of chip profit. System companies are the primary beneficiaries of SoC integration. They can design just what they need for their specialized product and they are better able to absorb the cost of the design.
During the protracted standardization cycle, many system companies de-emphasized internal SoC design expertise. With customization comes a renewed importance for front-end tools that support SoC architectural exploration and specification, rapid prototyping, and integration of all forms of semiconductor IP. IP integration aided by on-chip networks and their accompanying development tools especially accelerates the process of SoC design. Finally, this shift to customization presents an opportunity for semiconductor manufacturers and IP suppliers to provide their customers with higher value-added design services while system companies rebuild their internal expertise and focus their chip design investment on the parts of the SoC that add the most differentiation.