The CTO interview with Ambiq’s Scott Hanson

The CTO interview with Ambiq’s Scott Hanson

Interviews |
By Peter Clarke

Hanson, a graduate from the University of Michigan, founded Ambiq Micro Inc. (Austin, Texas) back in 2010 on the simple premise that as power consumption scales with the square of voltage: the lower the operational voltage, the lower the power consumption of any given circuit.

So here is a little background: conventionally MOS transistors are designed to operate in the strong inversion zone where voltage and current scale linearly. This allows transistors to switch faster but it is possible to take transistor operation close to and even below the threshold voltage. So, for manufacturing processes at around 40nm or 28nm this would mean operating circuits at around 0.3V to 0.5V rather than the conventional case of 1.0V or more.

But this simple idea is not so simple to implement because different types of circuit are more or less tolerant of voltage scaling and may need significant modification. Also, the digital circuit clock frequency must be scaled back with voltage, which reduces the peak performance. But, generally, for energy efficiency sub- or near-threshold operation has been recognized to be good thing and Ambiq was one of a few companies that started to consider it in the early part of the previous decade.

We asked Hanson, if sub-threshold operation is a good idea, why is it still an outlier activity and most MCUs sold operate at higher voltages with inferior power efficiency.

“Most of our competition would rather have the foundry solve their problems for them,” said Hanson. While it is the case that foundries have been characterizing manufacturing processes for lower voltages than they conventionally did, they are not going sub- or even near-threshold because they want to offer a relatively unified support in terms of foundational intellectual property cores.

Next: Energy efficiency

“Dialling down the voltage is not simple. It is not the case of just using a different PDK [physical design kit]. Ambiq redesigns everything, every analog block, voltage regulation. This keeps most [MCU] companies away.” But it also means that there are MCU customers, those driven by power consumption limitations, who are eager to use Ambiq’s offering, he said. The only other way to get improved energy efficiency is by physical scaling.

“Moore’s Law is not good enough for most customers. Unlike those making processors for servers and smartphones, our customers can’t pay for 10nm and 7nm chips. They are at the 40nm to 22nm nodes,” said Hanson.

Ambiq works closely with foundry supplier TSMC and over Ambiq’s 11-year existence has moved down through TSMC’s manufacturing nodes, but behind the leading edge. From 180nm to 90nm, to 40nm and on to 22nm. “We are preparing to go to 12nm,” said Hanson. This is a move that could make a significant difference to Ambiq’s application reach starting to take the company beyond the battery-operated applications that have been its strength to-date.

100 million units

Ambiq claims to have commercial traction and to be shipping million units per month and more than 100 million units in total.

Ambiq’s milestones to 100 million units shipped. Source: Ambiq Micro Inc.

“I can’t say precise numbers but revenue has been doubling each year recently,” said Hanson. “Our big problem is meeting demand,” he added, touching on something that is affecting the semiconductor supply chain across almost all application sectors. “TSMC is a great partner for us but growth is being affected by supply limitations,” Hanson said.

As Ambiq heads into FinFET manufacturing processes it begs the question whether sub-threshold voltage operation could be applied to fully-depleted silicon-on-insulator (FDSOI) manufacturing processes.

Hanson replied: “FDSOI processes are indeed an interesting option for sub-threshold and near-threshold operations. Various characteristics, including the steep sub-threshold slope, are very attractive for low voltage operation. FinFET devices offer many of the same benefits as well.  You should expect to see Ambiq using some of these more advanced processes in the near future. Note that different transistor types can help enable sub-threshold or near-threshold operation, but they are only one piece of the puzzle. They still need to be paired with proper circuit designs, chip architectures, test methodologies, etc., to ensure robust and efficient operation.”

Next: Three markets and RISC-V

So how did Ambiq get to 100 million units shipped while escaping the notice of many MCU buyers?

“Wearables has been a ‘beach-head’ market for us. It’s a market that really cares about power,” said Hanson. “No names allowed, but most of the top wearables companies are using our processors today. I should say we dominate and have shipped more into that market than any other MCU supplier.”

And having secured the beach-head the company is now moving into adjacent markets and with that comes a significant workload.

Hanson said Ambiq now has three application areas of focus: smart personal devices, smart home devices, and smart industrial devices. This will continue to be mainly battery-operated but not exclusively and they cover everything from headphones, wearables and sports equipment, up to climate control, sensor systems, utility meters and security systems.

“We need to grow the application knowledge base, reference designs, and put more FAEs [field application engineers] in front of customers. Whenever we appoint an FAE they are immediately saturated with demand,” said Hanson. “We also need to rapidly expand our portfolio of products with different feature sets and cost points.

A look into the product catalogs of broad-line MCU suppliers reveals hundreds of MCU variants. “We have multiple chip design teams now,” said Hanson, but clearly this is something that Ambiq must choose to address in a strategic way.

Almost since its formation and early collaboration with ARM, Ambiq has been exclusively a Cortex-M microcontroller supplier. We asked Hanson if Ambiq would be adding the RISC-V processor architecture to its portfolio.

“We’re interested in RISC-V. It’s not a move we’ve made yet. The first use of proprietary cores will likely be where the customer doesn’t need to write software, such as a radio protocol controller,” Hanson said. “You will see non-ARM cores pop up in our MCUs. We included a GPU core in the Apollo4. We have licensed the HiFi-5 DSP core from Tensilica. The HiFi-5 is a vector DSP core that is good for both signal processing and inference for AI applications,” said Hanson (see Ambiq launches Apollo4 SoC on 22ULL and Ambiq preps processor platform with NN support ). Hanson said there is now a lot of focus on inference processing for sensor interface circuits.

Next: IPO

Circling back to the supply restrictions, we asked if Ambiq needs to work with a second foundry, despite the large amount of work that would entail. Any development of sub-threshold FDSOI, hinted at earlier in the interview, would probably involve working with either Globalfoundries or Samsung or both, as TSMC does not offer any FDSOI manufacturing processes and seems unlikely to start.

“We have a tight partnership with TSMC. We are thinking about how to manage our risk. If we went public we would have a responsibility to the broader shareholders,” Hanson said.

So, is Ambiq preparing for an initial public offering of shares or looking to go public by way of a reverse take-over of a SPAC vehicle?  “An IPO is a very near-term possibility. There’s a real interest. I think we would do very well as a public company,” Hanson said

And given the number of companies relocating their primary domicile (Ireland draws startups to relocate from California) will Ambiq continue to be headquartered in Austin, Texas. Reportedly pressure is coming from China generally to disengage with US-based companies, creating sales and investment problems (see CEO interview: Palma Ceia’s Jewell on a pivot to fabless).

“We like Austin. There’s a flow of talent moving here. Bringing their equity from Silicon Valley. There’s a good entrepreneurial scene. Half of our company is in the US and half split between Taiwan and China. The parent company is US-based but we are a multinational company. We have to address both the US and China,” Hanson concluded.

Related links and articles:

News articles:

Ambiq offers ‘Always-on-Voice’ dev board

Ambiq gets Citizen watch design win

Ambiq launches Apollo4 SoC on 22ULL

Ambiq preps processor platform with NN support

Minima, ARM apply ‘real-time’ voltage scaling to Cortex-M3

CEO interview: Palma Ceia’s Jewell on a pivot to fabless

Ireland draws startups to relocate from California

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles