Chip Design and Cloud Computing: a Perfect Storm
Session Details
Session Abstract
Chip design is at the core of consumer electronics: smart phones, tablets, computers, HDTV, networking, etc. It is a $300B market that has been following Moore’s Law for 40 years: double the transistors density every two years.
This exponential complexity growth is reaching the limit of conventional computing frameworks. A single logical simulation of a chip exceeds 20 CPU years. Physical verification, which checks that a chip meets strict manufacturability rules, is reaching 100 CPU years. Many of these simulations are needed throughout a chip design cycle.
Chip design depends on the EDA industry (Electronic Design Automation), which provides the software to design, test, manufacture, and verify VLSI circuits. EDA sells TBLs (Time Based Licenses), usually for durations of 1 to 3 years, at prices ranging from $5,000 to more than $300,000. Because of TBLs, it can be very expensive to exploit the parallelism of some tasks (e.g., simulation), since every node must acquire a license.
EDA also provides ready-to-use hardware functional blocks, called silicon IPs (Intellectual Properties), that are optimized for speed, area, or power consumption. As reflected by their name, silicon IPs must be protected from intruders looking for replicating their features. In general, chip makers are extremely sensitive to security, as innovative designs and features can make or break a company.
Last but not least, it is estimated that 30 to 40% of all EDA software use happens via pirated licenses.
Need for burst computing, rigid TBLs, extreme security needs for both vendors and customers, and a pervasive software piracy: we show how cloud computing is revolutionizing the chip design and EDA industries. We explain the move from computing farms and time-based software licenses, to elastic computing and per-pay-use business models. Although elastic computing is a benefit that nobody denies, the transition from TBL to pay-as-you-go poses complex problems in terms of software usage control and business models. Also the amount of data that needs to migrate in and out of the cloud can become a bottleneck. We show that a common cloud platform to deploy and use software for the lifetime of a project (usually several months) enables new benefits for the EDA vendors and their customers –monitoring, cost optimization, version control, flexible account, etc. We discuss the security aspects of the platform for both vendors and customers --software, IP, and design. We finally discuss the impact of this evolution on EDA software piracy. All these themes will be illustrated by real-life examples.
Speaker
