For this 3-part series, ASN spoke with Kelvin Low, senior director of marketing for Samsung Foundry and Axel Fischer, director of Samsung System LSI business in Europe about the company’s FD-SOI offering. Here in part 2, we’ll talk about design. (In part 1, we talked about Samsung’s technology readiness. In part 3, we’ll talk about the ecosystem.)
~ ~ ~
ASN: Let’s start by talking about value. What do you see as the key advantages of 28nm FD-SOI?
Kelvin Low: FD-SOI is wide-ranging. What I mean by this is for the designers, there are many design knobs available that you can use to achieve either high performance or ultra low power. That’s a an extremely valuable and important proposition. The wide dynamic performance-power range is achieved with FD-SOI’s body biasing ability. Though bulk technologies allow body biasing, it has a comparatively much narrower range.
Another key benefit is the super analog gain and properties of FD-SOI. I think moving forward, we’ll probably start to see more customers that are analog-centric. Later on, we’ll see this as one of the key value propositions of FD-SOI. Today, there’s still a lot of digital customers that we’re engaged with right now. The analog customers are still not yet aggressively migrating to [[more advanced]] technology nodes, but when they come, this will be an important distinction in FD-SOI vs. bulk.
Another important distinction not related to power-performance-area is the robustness of the reliability. This is a well-proven fact that FD-SOI is much more robust for soft-error immunity as compare to bulk. So anything that needs radiation protection (for example, military, aerospace – but those are not really the high-volumes), as well as automotive products, you’ll see value of better SER immunity as compared to bulk. Not just memory SER but logic SER. There are available design techniques to overcome / account for that. For example, if you design to overcome SER, you incur overhead in area for example. With FD-SOI, this is intrinsic, so you don’t need design tricks to suppress it.
ASN: When should designers consider using 28nm FD-SOI as opposed moving to 14nm FinFET or choosing another 28nm technology?
KL: By virtue of one being 28 and the other being 14, if you do need a lot of logic feature integration, or die-size reduction, 14nm will obviously become more necessary. If you just are looking for power savings, both 14nm FinFET and 28nm FD-SOI are fully depleted in nature, so both are able to operate with a lower power supply. So those are similarities. 14nm FinFET does provide higher performance compared to 28nm by virtue of how the process is constructed. Lastly, cost, which is related to the number of double-patterning layers – at 28nm, avoiding all the expensive double-patterning layers and 14nm having double-patterning being necessary for all the area scaling – that presents itself as a real difference. The end-product cost can also determine the choice of the technology selection.
Axel Fischer: The end-product cost, plus as well the investments from the customer side: the customer has to make a certain investment to develop the chip in terms of overall cost. If you look at photomask payment, NRE* and so on – this is weighting strongly, more and more as you go forward with advanced node technologies. There’s a set of customers that are feeling very comfortable to stay on the 28nm node.
KL: There are several 28nm flavors. There’s Poly-SiON, there’s HKMG, and there’s HKMG-FD-SOI. In terms of performance, there’s really a very clear distinction. In terms of power, you see a more radical power reduction with FD-SOI. In chip area scaling, I’d say roughly the same between HKMG and FD-SOI. This is dictated not so much by the transistor but by the overall design rules of the technology. So, 14nm is the higher cost point. 28nm is a much lower cost point, so overall a given budget that a customer has can determine whether 14nm is usable or otherwise. We have to sit down with the customer and really understand their needs. It’s not just trying to push one over the other solution. Based on their needs, we’ll make the proper recommendations.
ASN: Can designers get started today?
KL: We are moving FD-SOI discussions with customers to the next phase, which is to emphasize the design ecosystem readiness. So what we’ve been working on, and we really appreciate ST Micro’s support here, is to kick-start market adoption. We have access to ST Micro’s foundation library, and some of their foundation and basic IPs. Here, Samsung is distributing and supporting customers directly. They need to only work with us, and not with ST Micro. So they have access to the IP through us. We also provide design support, and we have additional IPs coming in to serve the customers from the traditional IP providers.
Many designers are new to body biasing. Fortunately, there are a couple of design partners that can help in this area. Synapse being one of them; Verisilicon another. Already, they have put in resources and plans and additional solutions to catalyze this market. In short, the PDK is available today, and the PDK supporting multi tools – Synopsys, Cadence and Mentor – are all available for download today. Libraries are also all available for download.
There’s nothing impeding designers from starting projects now. This is why we believe that 28FDSOI is the right node, because we are enabling the market to start projects today. If we start something else down the road, like a 14nm FD-SOI, for example, or something in between, the market will just say, hey, we like your transistor, we like your slides, but I have nothing to start my project on. So that is bad, because then it becomes a vicious cycle. We believe we have to enable 28nm designs now. Enable customers to bring actual products to the market. Eventually from there you can evolve 28 to something else.
ASN: Let’s talk some more about design considerations and body biasing, how it’s used and when.
KL: Both 14nm FinFET and 28nm FD-SOI are fully depleted. One unique technology value of fully-depleted architecture is the ability to operate the device at lower power supply. So power is the product of CV²/frequency. If you can operate this chip at lower power supply, you get significant dynamic power savings. FinFET does not have a body effect, so you cannot implement body biasing – it’s just not possible.
FD-SOI, on the other hand, has this extra knob – body biasing – that you can use. With reverse body bias (RBB), you can get much lower leakage power. If you want more performance, you can activate the FBB to get the necessary speed. Again, this is not possible with FinFET. So that will be one distinction. It depends on how you’re using your chip. It all depends on the system side, or even at the architecture side, how is it being considered already. If you’re already very comfortable using body biasing, then going to FinFET is a problem, because you’ve lost a knob. Some would rather not lose this knob because they see it as a huge advantage. That doesn’t mean you can’t design around it, it’s just different.
There are already users of body biasing for bulk. For customers that already use body biasing, this is nothing new. They’re pleased to now have the wider range, as opposed to the more narrow range for bulk.
AF: And probably going to FinFET is more disruptive for them. With FinFET, you have double-patterning considerations, etc. More capacitance to deal with.
ASN: Porting – does FD-SOI change the amount of time you have to budget for your port?
KL: If a customer already has products at 28nm, and they’re now planning the next product that has higher speed or better power consumption – they’re considering FinFET as one option, and now maybe the other option available is 28nm FD-SOI. The design learnings of going to FinFET are much more. So the port time will be longer than going to 28nm FD-SOI. We see customers hugely attracted because of this fact. Now they’re trying to make a choice. If it’s just a time-to-market constraint, sometimes FinFET doesn’t allow you to achieve that. If you have to tape out production in six months, you may have to use FD-SOI.
AF: Another key point for customers deciding to work with 28FDSOI is the fact that Samsung Foundry has joined the club. A few customers really hesitated on making the move to 28nm FD-SOI ST Micro is a very really advanced company, doing its own research and development, but the fact that the production capability was very limited has people shying away. Besides the technology, the presence and the engagement of Samsung is giving another boost to the acceptance.
KL: Yes, we’re recognized as a credible, high-volume manufacturing partner. That helps a lot.
~ ~ ~
*NRE = non-recurring engineering. In a fabless scenario, there are NRE for IP and design (engineering costs, up-front and royalty-based IP costs), NRE for masks and fabrication (mask costs, wafer prototype lots, tools costs, probe cards, loadboards and other one-time capital expenditures), and NRE for qualifications (ESD, latch-up and other industry-specific qualifications, as in automotives).
~ ~ ~
This is the second installment in ASN’s 3-part interview with Samsung on their 28nm FD-SOI foundry offering. If you missed the other parts, you can still read part 1 about technology readiness (click here), and part 3 on the ecosystem (click here).