Chiplets and IP reuse made easy: Smaller node ASICs are not always the best way forward

Chiplets are a relatively new trend in helping designers of large complex system design to enable a cost-effectively modular approach to designing with costly SoCs. I say relatively because proposals started back in 2016 when DARPA used it as part of its CHIPS (Common Heterogeneous Integration and IP Reuse Strategies) program.

To quote DARPA, “The monolithic nature of state-of-the-art SoCs is not always acceptable for Department of Defense (DoD) or other low-volume applications due to factors such as high initial prototype costs and requirements for alternative material sets.”

If you look at the economics of SoCs it’s easy to understand why chiplets are of increasing interest to a wide range of system designers – not just the likes of DARPA: moving from a 45 nm process to a 16 nm process more than doubles the cost/mm². Migrate again to a 7 nm process and costs / mm² double again – that’s 4x the cost per yielded mm².

Chiplets do sacrifice some space (around 10 percent) for the ‘chiplet architecture’ – i.e. the chip interconnects, but the overall benefits of chiplets on total system costs is substantial.

The fundamental benefits of a chiplet approach are: low cost and easy way to repurpose the design and enable variants, with some die-to-die interconnect, a range of third party chips can be combined into a package. One of the most important contributing factors enabling the chiplet approach to system level design is today’s advanced packaging technologies.

Many of the benefits and how chiplet design works may sound rather familiar because it’s what Thalia Design and others in the IP Reuse industry have been enabling for decades. In fact, we see huge potential in chiplets and, done correctly, a strategy of reusable chiplets will present enormous time and cost savings. Time to market, ease of design / migration to new processes, familiarity with the design…all of the benefits Thalia’s AMALIA (TM) IP Reuse platform enables.

When approaching the chiplet design, you can create a portfolio of IP for dedicated functions, even before committing the design to silicon. This makes your chiplet design both technology-agnostic and fab-agnostic.

Thalia’s AMALIA Circuit and IP reuse platform gives you that flexibility. Thalia methodology ports your existing chiplets to new technology nodes or fabs in the shortest possible time.

Commercial considerations can be taken into account during the implementation. For example, if the final solution needs to use a specific foundry, for strategical reasons, the chiplets can be quickly and easily ported to that specific foundry and even technology node.

That raises the levels of flexibility in complex system-level implementations and also gives customers the maximum flexibility in defining their system-in-package solution.

The AMALIA platform only needs the chiplet source technology database and PDK, plus the target technology PDK. AMALIA can also help to create a complete chiplet portfolio. Thalia IP reuse suite of products can handle the complete design, reuse or migration of the chiplet in practically every fab or technology node.

Thalia’s partner network also provides the productization of chiplets in specific foundries, down to 7nm technology nodes.

If you are planning your own chiplet strategy, get in touch with us to find out how AMALIA platform can help to further simplify the IP reuse.

Thalia Design Automation partners with Sofics to enhance offering for analog circuit and IP reuse

Partnership enhances market-leading Technology Analyzer with robust hardened I/Os and ESD protection options

Cologne, Germany, 16 June 2021 – Thalia Design Automation Ltd., provider of analog and mixed-signal circuit IP reuse platform, today announced a new partnership with Sofics, a leading provider of analog I/Os, specialty digital I/Os and ESD protection.

Thalia’s Technology Analyzer, part of its AMALIA platform, helps major IP houses and integrated circuit design firms determine whether or not IP is fit for purpose or suitable for cost-effective migration. This partnership means Sofics’ specialist IP portfolio will now be included in the recommended target technology candidates.

“This announcement is about the coming together of two market-leading solutions,” said Sowmyan Rajagopalan, Thalia Design Automation CTO. “This latest iteration of our Technology Analyzer uses machine-learning to rapidly compare the source and target process technologies, enabling business case and commercial modelling to be undertaken for our clients. They are then able to assess migration value before committing resource. Adding Sofics’ ESD protection and other I/Os into the target options will enable solutions to be found and implemented much more rapidly.”

“IP houses want to optimize the value of existing assets and extend portfolios through IP reuse,” said Koen Verhaege, CEO of Sofics. “Collectively, this is a solution that mitigates risk for customers when transitioning to new technology nodes. Sofics’ global reach means that we can bring a world-class technology solution to IP and IC firms.”

The AMALIA Technology Analyzer is an intuitive solution that addresses a comprehensive array of first and second order effects including FT, gm/id, Vdsat, Vt mismatches, corners, Monte Carlo mismatch impact and many others. The software integrates into several EDA design frameworks and can be customized to specific design flows.

Sofics leading I/O and ESD foundry independent IPs are adaptable to various technologies like high & low voltage, BCD, CMOS, SOI and FinFET. More than 100 fabless companies use Sofics’ solutions to enable higher performance, higher robustness while reducing design time and cost of SoC design.

Find out more about the AMALIA Technology Analyzer.

AI: when is it ‘really’ intelligent?

Artificial Intelligence is quite obviously a buzzword which attracts significant marketing hype – that has been the case for a decade at least. There are countless number of high-profile examples where AI is used to simply describe an automated (usually software-enabled) routine. Good examples of this can be seen in Facebook’s use of AI: filters that simply track and flag keywords, or images, that break a set of human-defined rules. The fact of the large number of false positives they ‘capture’ demonstrates that, while these programs may be artificial, they’re not always intelligent as we humans would define it. They are, more often than not, just forms of computational automation.

Don’t get me wrong, computational automation can be beneficial, it can speed things up and save significant time, hence money. But it does not add skills, nor does it bring added, intelligent value to a design team – which is what we’re trying to do for our customers. Given the wide and potentially misleading use of the term, there is no doubt that when we chose to use AI to refer to the capabilities of our AMALIA Design Enabler we really had to pause and check we were being honest with ourselves, and with our customers.

AMALIA Design Enabler passes the acid test. We threw problems at the system, asking it to find solutions to problems – looking for answers we didn’t know existed. In the video, you can see a perfect example of its application and the benefits of a truly-AI solution for supporting design problems: helping the designer to compare and assess alternative components to resolve a technical issue with the system. In this case, during a process migration, a very low current voltage regulator was taking far too long to achieve a zero temperature coefficient state… to compare the options and find a solution just in the small circuit in question could have taken the design team several weeks.

By entering the requirements into AMALIA Design Enabler, along with the options available, the Design Enabler AI algorithm was able to find its own way to the answer in relatively few steps and a very short amount of time: reaching the answer in just 40 steps, in spite of there being thousands of potential variables.

The reason we’re confident in calling the Design Enabler true-AI, is exactly that: the system is learning as it goes, changing direction based on initial findings and zeroing-in on the correct solution: it is not simply observing, calculating, and running every possible scenario before ‘happening’ upon the correct answer by brute force and computational power.

So, AI is dead. Long live AI! We believe that true-AI does have a place in design automation. Selecting the appropriate components and enabling appropriate IP reuse in process migration can be a complex, time-consuming task – but when it’s done well, it significantly improves design profitability and system performance.