Can AI Alter The Burgeoning Design Cost Trend?

Everyone in the semiconductor design arena has experienced or at least observed the impact of increasing costs for complex SoC silicon. Semico’s recently released report entitled “Silicon and Software Design Cost Analysis” reveals the cost associated with a first time design effort for a high-end, advanced performance multicore SoC using 7nm process technology can top $195M for both the silicon and the software. The software effort accounts for the largest share of the cost; however, the silicon design costs have gone up significantly.

From 2008 to 2022, the number of most expensive first time design efforts average 2.6% of all designs. That number increases slightly towards the end of the timeframe. Semico believes this is due to the increasing number of advanced designs such as the ones that incorporate artificial intelligence functions.

In general, design costs for advanced SoCs are increasing at a 36.1% CAGR. This growth rate is calculated by process node. It started from the 90nm node and is projected through the 3nm node; it assumes that all design parameters are maxed out in a first-time effort.

Complexity levels continue to increase and Semico forecasts the average number of IP blocks for all SoC types will be 157 in 2018. The average for advanced SoCs is much higher at 205. Basic SoCs have a lower IP content compared to the other SoC types and tend to lower the average number of IP blocks for the overall industry. Semico expects this to change over time as basic SoCs start to contain higher numbers of IP blocks due to increasing market requirements for IoT solutions and the emergence of basic SoCs with AI functionality.

On average, waiting one year after the introduction of the latest geometry will produce a design that is roughly 30% less costly than doing a first effort when the new process technology first becomes available. But if your goal is to be the first to market what other options are available?  Can artificial intelligence improve the design landscape by getting more designs into the market and at a faster rate?

AI is being touted as a mechanism to provide some relief from the burgeoning SoC design costs.  As EDA tools with AI functionality are used in the design effort, designers will be presented with the potential of finishing a design in less time or including more functionality in the same amount of time. A time savings equates to a reduction in the design cost while preserving the same level of complexity and functionality. It is difficult to estimate exactly how much time will be saved in this manner, but a reasonable estimate could be on the order of 20% – 25%.

Designers could opt to use the new-found efficiency to include more complexity and functionality in the same amount of time as their original plan called for and forgo any potential cost savings. The choice between saving time or adding more functionality will be determined on a design-by-design basis. It will truly be up to the system architect, individual designer, the company product strategy and the customer’s system requirements. In the past, most companies added more functionality.

Semico expects AI functionality to become a ‘check-list’ item in most electronic systems and in EDA tools as well. For the EDA market, the AI functions will be used as an aid to the designer when analyzing thousands of verification runs or in SPICE modeling. Many of these analysis tasks are already automated but require human intervention to wade through the myriad of results delivered by the simulations. This process can be further augmented to perform analysis on the raw results to boil down the mountain of data so it can be much more useful. AI can filter the results to a more meaningful set of data points for the designer thereby saving time, releasing design resources for more important tasks. The first such tools entered the market in 2017 and will continue to be introduced in 2018 and beyond.

There are two important trends that Semico believes are going to be major factors in the semiconductor industry in the coming years. Not only will AI functions increase the need for electronics that require semiconductors with AI functionality, AI will also be used in the semiconductor design process and throughout the semiconductor supply chain, from mask making to manufacturing to the creation of IP and software.

Inclusion of AI functionality into EDA tools will not replace designers, it will augment the capabilities of designers to make better decisions at a faster rate. It will provide more headroom to designers and allow for the redistribution of scarce resources to be focused on other pressing issues in the design process. The same will be true in the other areas mentioned: mask making, manufacturing, and creation of IP and software. It will allow for more robust testing and monitoring functions while performing these functions faster and with more reliability. These factors can be included into the design effort and help produce better designs and products for the industry.

Semico believes the real key to all these issues is the continued growth in unit demand and the increasing market requirements for more functionality and richer feature sets in the electronic devices we have all come to depend on in our society today. The high device demand and revenue growth pushes designers, researchers and companies all around the world to continue to look for solutions to current bottlenecks in design and manufacturing. Even though design costs for silicon and software continue to rise, the advent of new tools, functionality and capabilities will act to moderate these increases somewhat over the long term and give designers new tools to positively affect the rise of design costs.

The good news is that the industry continues to find new solutions and improve existing solutions to address the issues we face in continuing to create cutting-edge silicon. Ingenuity and inspiration have great value, and they have found an important place in today’s semiconductor industry. The future continues to be very bright indeed!