The past year has seen a resurgent interest in 200mm fabrication. In this paper, I will discuss why this is and answer the question, "Can 200mm fabs have a profitable future?"
I will also share some of my ideas to maximize profitable growth for mature “More than Moore” foundries. These ideas were shaped by my experience at Globalfoundries and managing several fabless companies.
Why the renewed interest in 200mm?
From my experience at Globalfoundries, I realized that leading edge process technology was becoming less and less affordable. This shaped my “Law of Process Scaling Economics.”
Simply put, as transistor scaling advances, development costs climb dramatically, decreasing the number of customers who can afford the technology. These costs are well understood and documented such as fab construction, semi equipment, triple & quadruple patterning, etc. Less understood and recognized are the costs of intellectual property such as cores, memory, interconnect and the associated validation costs.
As a result, IBS estimates the product revenue required to justify a leading-edge design will skyrocket from greater than $300 million for 28nm to billions of dollars at 10nm.
While development costs have been climbing, another megatrend has emerged to make leading edge economics even more challenging. For most of the semiconductor industry’s history there has been a single market driver.
- In the 1950s, it was military and aerospace;
- The 60s, mainframe computing;
- The 70s, minicomputers;
- The 80s, personal computers;
- The 90s, networking;
- During the past 15 years, mobile has been the main driver.
However, for the first time, semiconductor growth is now not driven by just one main end market. Today’s market is driven by the Internet of Things, which is not a single market but a myriad of diverse end markets and applications.
IoT is really the “siliconization of everything.” Most of these devices will have a modest volume in comparison to the market drivers of the past. Very few IoT products will have 1 billion unit or even a 100 million unit potential. Rather, this is an age of 10,000 different devices which might sell 100,000 units. As a result, there are few devices that have the volume to justify the most advanced technology. IBS estimates that there will only be a handful of IoT products that will sell more than 10 million units in the year 2020.
Clearly, there is a disconnect between the IoT market driver and leading-edge costs.
But happily, many IoT devices are predominately analog or mixed-signal and do not require nor benefit from the most advanced technology. They are the “More than Moore” functions such as sensors, power, human interface or RF. These functions do not scale downward and do not like the lower thresholds or currents in the most advanced nodes. For this reason, 130nm had the largest number of design starts in 2015; 2x the next nearest node, 180nm While other nodes are climbing, such as 65nm, 40nm and 28nm, for the next several years, 130nm will remain the most popular technology.
(Source: IBS, SEMI, GSA)
During this transition to IoT as a major driver of semiconductor growth, the number of 200mm fabs decreased substantially. In 2007, there were 199 200mm fabs worldwide and by 2015 this had dropped to just 178. There were two major reasons for this decrease:
- The continuing shift by many semiconductor companies from an integrated device manufacturer (IDM) model to a fabless/fablite–foundry model.
- The unprecedented number of industry mergers and acquisitions. Many of these deals either assumed a fabless model or a consolidation of captive manufacturing to reduce operating expenses.
The low point in number of 200mm fabs worldwide came in 2015. This resulted in many foundries reporting their 200mm capacity being 100 percent utilized and on allocation throughout 2015 and 2016.
NEXT PAGE: 200mm Making a Comeback