It's been a wild ride at Marvell, and it's not over yet.
Neil Kim is a booster shot of positive practical engineering energy for Marvel, which is currently undergoing a major surgical makeover. The former head of Broadcom’s central engineering group spent an hour with EE Times sharing his views at the end of his first week on the job as Marvell’s new chief technologist.
It’s been a wild ride at Marvell, and it’s not over yet. A year ago, Marvell founders—CEO Sehat Sutardja and President Weili Dai, his wife—resigned as the company’s finances tumbled amid investigations of accounting practices. Revenues and profits had fallen from $3.6 billion and $435 million in 2015, respectively, to $2.6 billion with an $811 million loss a year later.
Since then, the company has essentially replaced its entire board and top tier of management. Last May, Rick Hill, former chief executive of Novellus Systems, became Marvell’s chairman. In June, Matthew Murphy, a senior executive from Maxim Integrated, was named chief executive. In less than a year, Murphy has hired half dozen execs under him, with Kim being the most recent, and the third to come from rival Broadcom.
The financial free-fall stabilised in Marvell’s last fiscal year to revenues of $2.3 billion and a $21 million profit. The company has 4,600 employees, down nearly 40% from its peak of 7,300 in 2014. Current management estimated by October 2017 900 will have left, and it has earmarked an undisclosed number of products it aims to cancel or sell off.
Murphy has indicated that he aims to complete the restructuring by October. Even when he is done, Marvell still faces a number of strategic challenges.
While it has been focused on rebuilding itself, the semiconductor industry around it has gone through a historic consolidation. Marvell’s top competitors in embedded processors, Ethernet and Wi-Fi have beefed up for the next rounds of battle amid slowing market growth. Broadcom merged with Avago, and Qualcomm is set to gobble up NXP, enabling new levels of chip integration, savings, and economies of scale.
About half of Marvell’s current business is in a sunset sector—controllers for hard disk drives. It’s a business that depends on a few big customers.
Marvell’s second largest customer after Western Digital—Toshiba’s electronics group—is now up for sale. Its third biggest customer, WinTech, is a relatively small memory module maker.
Former CEO Sutardja was himself an engineer at heart. He was known for getting deeply involved in projects such as MoChi, a silicon interconnect that he described in a 2015 ISSCC keynote, one of his last major public talks.
Neil Kim spent the last 15 years of his career building and overseeing Broadcom’s central engineering group. It was known for being a disciplined practitioner of silicon design and reuse, a skill that the company relied on repeatedly as it grew by acquisition.
Kim retired from Broadcom last year just before Avago closed the deal to buy the company in one of the largest of many recent large mergers. We spoke to Kim after his first week at Marvell’s Silicon Valley headquarters.
Tell us the story of how you came to Marvell.
I retired from Broadcom. I wasn’t really looking. One of the Marvell board members—Robert Switz, whom I knew from Broadcom—called and asked if I could meet Matt [Murphy]. I thought it was good to meet him as a courtesy, and it’s good to know people like Matt.
I knew something of what was going on at Marvell. Matt talked about what he was doing with restructuring, and he had done quite a bit of work on it already. He came across as a nice person, very straightforward, and I liked what he was doing.
It’s a big change. All of the top staff is new and there’s a lot of enthusiasm. I know restructuring is a very difficult thing, but they are accomplishing things, getting out of some markets that were highly competitive and low-margin. Marvell got out of consumer products for TVs and cellular, and I had experience when Broadcom got out of cellular as well.
Now Marvell wants to focus on cloud computing and infrastructure, products that are at a higher margin and growth rate than the average of 3% for semiconductors overall.
I guess I struck a chord with [Murphy]. I joined for multiple reasons. I had always known Marvell as a competitor and had tremendous respect for the engineering team. With the restructuring, I thought that the future was bright for growth and profitability, and we have the portfolio to go after the market.
Were you a customer of Marvell when you were head of engineering at Western Digital?
I wasn’t a direct customer. We designed our own hard-disk SoCs. We did read channel stuff back in the day—not for our premium drives, but for cost reduction. I met Sehat and [his brother] Pantas when they just began Marvell [in 1995]. I don’t believe that they had a hard-drive product at that time. WD engaged with Marvell on read channels shortly after I left to go to Broadcom.
What was your experience like working on design and reuse with acquired companies at Broadcom? I got to see a lot of companies under the covers—small and not so small. I was in charge of digital IP sharing and IP development. There were times when I thought we could do much better, but having seen how others shared IP, I thought we had it down pretty well. You can always increase your efficiency.
Many companies that we acquired cooperated pretty well with the exception of very few cases. Centralised engineering was pretty large, so we had momentum and there was not a huge amount of war stories.
Were there a lot of concerns around the Avago acquisition?
After I left, I had multiple lunches with people I had worked with.
What lessons that you learned at Broadcom will you bring to Marvell?
I encouraged collaboration and coordination across business units. That’s something every company can do better. I want to be the advocate for our engineering community. Marvell already has a strong culture of innovation and execution. I want to embrace and enhance it with continuous training so people can sharpen their execution, streamlining processes and putting in new ones.
I was told that Marvell has had some problems getting first-silicon into production with products such as Bobcat.
I want to see the history of how we are executing. I’m presuming that we’re executing well, but I have not had a chance to look at the history yet. I am told that we have good IP checkout, watermarking and repositories. I have heard some things from the supplier side but not the customer side yet.
I have been in a couple of design reviews and roundtables with a couple of organisations so far. Our engineers are very capable, succinct and focused. I was pretty impressed, and even the roundtable discussions were very friendly.
Marvell’s R&D budget hovered around a billion dollars a year for several years. What’s the new target?
In the past, semiconductor companies in system SoCs typically spent about 25% to 30% of revenue on R&D. I’m now seeing a good portion of these SoC companies spending 15% to 20%. The trend has been to the mid-teens and the high-teens on the high end.
A number of analog companies are investing below 10% and are tremendously focused on profitability. Analog has matured a lot more than SoCs.
How will you diversify out of hard drives?
There is a lot of growth ahead in storage. The hard drive market is flattening and declining slightly over time. We are a leader, but there is still market share to be had in that area.
We also can grow in cloud storage applications and go into higher-value sockets in hard drives. There’s also an opportunity to engage in ASICs as well as merchant chip markets.
Networking is another growth area with a 10G Ethernet transition on many campuses. I’ve already seen some of the things we are doing in architecture that look very interesting.
We also are a leader in chips for wireless access points for the enterprise. And automotive is becoming a multi-billion-dollar market.
In automotive, are you focusing in part on your Ethernet chips?
Marvell was early in experimenting with ARM servers. Is that no longer a target?
We have multiple ARM core processors, but we’re not going after the server market.
What do you think will happen with MoChi, the interconnect that Sutardja pioneered?
It’s a cool concept, developing a modularized approach and packetised interface for time-to-market and flexibility. That can save cost and time. There are apps in which this can play well. It’s cool stuff that can be used to talk to multiple chips or IPs. There are three product areas where people are employing it. I have a meeting on it this afternoon.
Overall, what’s the job of the CTO?
The CTO’s role is to be an advocate for our engineering community at Marvell. I want to listen to their achievements and concerns and help them through transitions and be a coach, and I plan to grow with them—I’m willing to learn new things.
Keeping morale up and retaining our best people will be one of our tasks. Marvell is still an exciting place; we’re aligned with growth in storage and communications. We have a strong balance sheet and we generate a tremendous amount of cash. There’s an opportunity to grow with the company and have a culture of innovation.
What do you think about moving from southern California to Silicon Valley?
I told our HR vice president that my first impression is that people here are very friendly. I sometimes had an impression that this was a cutthroat valley, but I don’t find that true at all.
First published by EE Times U.S.