2 Discussion and 1 Research Paper

131

The network will truly be the computer.1

If you happen to make a wireless phone call to China, it may

be routed through a microprocessor chip produced by Raza

Microelectronics Inc. Although RMI (later merged with NetLogic

Microsystems) only began supplying integrated circuits in mid-

2005, its microprocessors now power some of the most advanced

communications network equipment in the world, and it counts

the leading Chinese equipment suppliers among its customers.

It has become the acknowledged technology leader in its chosen

market.

How a Silicon Valley startup developed the world’s most

sophisticated network microprocessor is a remarkable story. It is also

especially relevant to our theme of global entrepreneurship, because

RMI’s success hinged as much on its understanding of international

markets as on its technical expertise.

RMI’s management realized early on that overseas markets

were more open to network chip innovations from a startup than the

domestic US equipment manufacturers, who had well-established

chip suppliers with long-standing reputations. So the company

decided that its primary sales target should be China, the world’s

fastest-growing network equipment maker.

This may seem like an odd choice. China has become the

world’s factory. Its trade surplus with the US has risen from $202

billion in 2005 ($243 billion in exports vs. $41 billion in imports) to

5 Speeding voice and data traffic worldwide: Network microprocessors from RMI

1 http://googlesystemblogspot.com/2006/11/network-will-truly-be-compter.html, accessed November 4, 2011.

C o p y r i g h t 2 0 1 2 . C a m b r i d g e U n i v e r s i t y P r e s s .

A l l r i g h t s r e s e r v e d . M a y n o t b e r e p r o d u c e d i n a n y f o r m w i t h o u t p e r m i s s i o n f r o m t h e p u b l i s h e r , e x c e p t f a i r u s e s p e r m i t t e d u n d e r U . S . o r a p p l i c a b l e c o p y r i g h t l a w .

EBSCO Publishing : eBook Collection (EBSCOhost) - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS AN: 465765 ; Kressel, Henry, Lento, Thomas V..; Entrepreneurship in the Global Economy : Engine for Economic Growth Account: s8501869.main.ehost

Speeding voice a nd data tr a ffic wor ldw ide132

$273 billion in 2010 ($365 billion vs. $92 billion).2 American busi-

nesspeople and government officials routinely complain that, while

China floods the US with manufactured goods, the Chinese market

is closed to American products. But RMI’s success in China proves

that its market is very open to innovative products no matter where

they originate.

RMI also embodies another common characteristic of the suc-

cessful technology startups surveyed in this book: it was built by

immigrants. Its founder and first CEO, Atiq Raza, was born in Pakistan,

while his successor, Behrooz Abdi, is a native of Iran. The drive to suc-

ceed, tolerance for risk, and comfort with international dealings that

marks the educated immigrant community must be counted as an

important element in the company’s growth and success.

However, while global perspective and immigrant origins cer-

tainly help, they aren’t the whole story. Ultimately every success-

ful startup enterprise is built on the same foundation: recognizing

and filling an unmet market need. In RMI’s case, it was an urgent

demand for ever-higher speeds in network communications. This is

the story of how the company met that need through technical break-

throughs, business savvy, and the flexibility to adapt to changes in

the business environment.

Information autobahn

The demand for more speed in communications networks has been

building for decades, and shows no sign of abating. It really started

to ramp up about thirty years ago, when the Internet was just getting

started. At that time only large corporations could afford digital net-

work connections, and they ran over dedicated lines that were closed

to everyone else.

Smaller organizations and individuals had to make do

with analog telephone lines, using devices called modems (for

2 www.census.gov/foreign-trade/balance/c5700.html#2010, accessed September 9, 2011.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Infor m ation autoba hn 133

modulation/demodulation) to convert digital data to and from ana-

log transport modes. Early modems could handle a maximum of 300

bps (bits per second) of data. Eventually modems reached top speeds

of 56 kbps (thousands of bits per second). This was the equivalent of

about three typewritten pages per second.3

Today’s data networks are end-to-end digital. Comparing ana-

log modem speeds to what can be achieved in the digital realm is

like matching a slow stroll on a suburban sidewalk to a Grand Prix

race car on the autobahn. Broadband wired Internet connections

can deliver data at up to 100 Mbps (millions of bits per second) –

or 120 copies of the complete works of Shakespeare every minute.

Wireless is slower, but even the latest cell phones are capable of up

to 10 Mbps.

Digital data networks emerged on a global scale in the 1990s,

fueled by the rise of the Internet and its enormous data-handling and

security requirements. As people demanded ever more data-inten-

sive digital applications on their personal computers and wireless

handsets, the pressure to provide faster transmission speeds showed

no sign of lessening.

RMI was founded precisely to meet this demand for faster data

traffic. But providing higher data speeds isn’t as simple as stepping on

an accelerator. In order to appreciate the magnitude of the achieve-

ment, it is necessary to understand the basics of modern networks

and the hardware that runs them.

Routing bits and packets

The speed, flexibility, and reliability of digital communications,

from the simplest text email to video on the Web, are a direct out-

come of their “packetized” network structure.

Analog networks carry a continuous stream of information

that must be kept intact from origin to end-point. Digital networks

3 See www2.sims.berkeley.edu/research/projects/how-much-info/datapowers.html (accessed September 9, 2011) for a handy table by Roy Williams that matches various quantities of digital data with everyday forms of information.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide134

are built on the radical concept that you can break up information

into smaller units, called packets, and send them as separate com-

ponents, to be reassembled into a single communication at their

destination.

This structure has two main benefits: the network can carry

many individual communications at the same time; and individual

portions of each message (the packets) can travel different routes to

their destination without destroying message integrity.

It may take millions of packets to contain the content of a

single data transmission. Each packet has two parts: the payload, or

data that has to delivered, and a header that identifies the packet’s

origin, contents, and destination.

The header is the crucial part of the message from the net-

work’s point of view – the payload is not so important. In fact, header

instructions can account for as much as half of a packet’s total bits.

Headers contain more than simple routing information – much

more. They identify the order in which packets are to be reassem-

bled, and ensure the packets are treated properly as they travel to

their destination. When the message includes voice and video, for

example, headers tell the network to treat the packets in such a way

that the contents play back smoothly at their destination.

When packets in a single transmission travel along different

routes to their destination, they may arrive out of sequence. This

is because the computers that direct the data traffic along the way,

appropriately called routers, are independent engines, giving the net-

work built-in redundancy. If one segment goes down, or a router is

busy, packets can be rerouted to their destination through alternate

paths.

At the end of the journey, a router must reassemble the mes-

sage in the proper sequence. Since data travels over an optical fiber

cable at the speed of light, the biggest limiting factor in network

speed is clearly the routers that analyze headers and move the pack-

ets to their destination. In order to speed up the network, faster rout-

ers must be created.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Infor m ation autoba hn 135

Microprocessors: Breaking the router bottleneck

That’s where RMI’s focus on faster microprocessors comes into play.

A router is simply a special-purpose computer, and microprocessor

chips are the brains of computing systems. To make routers work

faster, the speed of their microprocessors must be increased.

Anyone familiar with electronic products tends to talk about

microprocessors as if they always existed. In fact, it was only in the

1970s that a single microprocessor began to replace the cluster of

integrated circuit chips on a circuit board that formed the central

processing units (CPUs) of previous computers.

The concept of a single chip designed to perform the core com-

puting operations originated at Intel. Its first commercial micro-

processor chip, dubbed the 4004, was introduced in 1971. The 4004

contained only 2100 transistors, but it had the same computing power

as the ENIAC computer of World War II vintage, which weighed 30

tons, occupied 3,000 cubic feet and used 18,000 vacuum tubes.4 (Of

course, this was before the invention of transistors.)

Gordon Moore, co-founder of Intel, saw the microprocessor as

an alternative to building custom chips for each computing system.

He described the device as “a broadly applicable, complex integrated

logic circuit that can be produced in huge volume, and hence [at] low

cost, and that utilizes the technology advantageously.”5

Moore’s idea was that all computing could be handled by soft-

ware running on a general-purpose microprocessor. Instead of hav-

ing custom hardware (chips) for each application, the microprocessor

would carry out all basic computational operations, with software

supplying application-specific functions. This would result in com-

puting systems that were both faster and cheaper, because these ver-

satile chips would be mass produced.

4 M. White, “25th anniversary for microprocessor,” Toronto Star, November 17, 1996.

5 G. E. Moore, “Microprocessors and integrated electronics technology,” Proceedings of the IEEE, vol.64 (1976), pp. 837–841.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide136

Moore’s vision of huge production volumes has been fulfilled

beyond anything he could have predicted. Today microprocessors are

everywhere, at all levels of performance. There are hundreds at work

for each human being in the world. As the speed and performance

of microprocessors have improved, and their cost has dropped, they

have found their way into everything from toys to washing machines

to large computers and communications systems.

In his famous “Moore’s Law,” the Intel founder also predicted

that the processing power of chips would double every two years

with no increase in price. This was based on the rate at which tran-

sistors were being shrunk in size, permitting more logic gates (and

higher performance) on each chip. Making transistors smaller thus

meant getting more speed for less money. But gate dimensions could

shrink only so far before they began to bump up against physical

constraints on performance. In particular, the power dissipated by

very-high-performance chips became excessive, and cooling them

increased the size and cost of the systems.

Fortunately, another path forward had emerged: multiple proc-

essors on a single chip working in parallel to process data. In these

configurations each processor core has its own resources, but shares

access to centralized control functions, including instructions and

data storage. Since the individual processors have significant auton-

omy in the execution of instructions, this arrangement achieves

much higher processing speeds even if the speeds of the individual

cores do not increase.

Parallel processing is not a new idea. It was the architecture of

choice in powerful mainframe computers that combined many dis-

crete processor chips. Designing a microprocessor chip containing a

group of suitably interconnected processor cores, however, is new. To

make such a device work called for not only extraordinary engineer-

ing skills but also the invention of new technology.

Multicore processors were just beginning to appear in the mar-

ket in 2002, the year RMI was founded. It was the challenge of design-

ing such devices to greatly speed up data networks that attracted

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

R ea dy for innovation 137

the company’s founding engineers. Their idea was to create a whole

new family of multicore microprocessors that would dramatically

increase the capacity of networks to meet the ever-growing demand

for higher-speed communications.

To do so, they had to design chips that processed the routing

information in packet headers in parallel rather than in the serial

manner of then-current single-core processors. This required the

invention of a novel chip architecture where multiple individual

processors were synchronized and shared timely access to infor-

mation needed to complete a computing task. This rapid access to

common information would enable a remarkable increase in packet

processing speed, eliminating the major bottleneck in data network

operation.

Designing the new chips was only the beginning. RMI’s cus-

tomers, the companies that built the routers, would have to learn

how to program the chips to meet their equipment objectives. For

this reason the microprocessors would have to be designed for versa-

tile programming. This opened the way for RMI’s customers to write

software that differentiated their network products by lower cost,

improved performance, and a superior ability to incorporate secur-

ity control. RMI’s founders were uniquely qualified to achieve these

goals.

Ready for innovation

Atiq Raza, the Pakistan-born founder of RMI, earned his degree in

physics and philosophy at the University of the Punjab, and then

moved to the US. He founded his first startup, NexGen, in 1988,

where he developed a general-purpose microprocessor designed to

compete with Intel’s top-of-the-line products. It was the first com-

pany to challenge Intel in the high-end microprocessor market.

Developing such a highly sophisticated product involved extra-

ordinary effort. NexGen had to avoid infringing Intel patents or copy-

rights, yet deliver products that were interoperable with Intel chips. It

took until 1995 for the company to develop a marketable product.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide138

At the same time AMD, a big chip manufacturer with a

license to serve as a second source for Intel processors, was also

committed to breaking the Intel monopoly. It attempted to design

its own microprocessors, but failed to come up with a competitive

product.

For AMD the obvious next move was to acquire NexGen.

AMD paid $850 million in AMD stock for the company, laying the

foundation for AMD to launch a series of successful chips to com-

pete with Intel. Raza joined AMD where, as president and chief oper-

ating officer, he led the microprocessor product line as it established

itself against Intel.

But not for long: always looking for the next challenge, Raza

left AMD in 1999 to become a venture capital investor. His idea was

to incubate new technology companies, and manage their growth

to the point where they would become attractive acquisition candi-

dates. Raza launched a number of companies but, when financing

startups proved difficult after the public market crash of 2000–2001,

he decided to refocus his efforts on building a single company tar-

geting specially designed microprocessor chips to improve data

traffic on networks. With the explosive growth of Internet traffic,

the amount of data carried was believed to be doubling every three

months, so better equipment, powered by faster microprocessors,

was urgently needed.

Raza founded RMI in 2002 specifically to launch a new gen-

eration of microprocessors that were designed to deliver a dramatic

increase in data network performance. This was clearly a growing

market that would be targeted by others, so RMI was off to a race to

be first with the best. In a first for the industry, these new products

would incorporate eight powerful, closely linked processor cores on

the same chip.

In addition to novel ideas, his competitive edge in starting

the company was a team of exceptional managers who had worked

with him before. These included Waqar Shah (head of operations),

Dr. Nazar Zaidi (head of product development), Dave Hass (product

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

R ea dy for innovation 139

architect), and Dr. Kai-Yeung (Sunny) Siu (head of sales and market-

ing for Asia).

Silicon Valley is the kind of place where talented engineers are

always willing to leave big companies to join a promising startup.

Raza’s reputation in Silicon Valley constituted a very high level of

promise. Everybody knew he had led the only successful challenge

to Intel’s monopoly in high-end microprocessors.

In a remarkably short period of time he assembled sixty of the

best microprocessor design engineers in the world. Their mission

was to produce the industry’s best single-chip data network micro-

processor. That’s the kind of challenge that gets the juices flowing

in that part of California. All that remained to be done was raising

enough venture capital to get to the finish line.

This is where Warburg Pincus came in. Our team was intro-

duced to Raza and his startup in 2002. Whether the proposed prod-

uct would find a ready market was not at issue. It was evident that

the rapidly increasing demand for data network traffic was spurring

a corresponding need for equipment to meet the demand.

What was startling was the ambition of the plan. There was no

precedent for the proposed product family in terms of performance

and value. Only a world-class talent had a hope of succeeding. What

convinced us to help fund the company (along with Benchmark

Capital) was Raza’s reputation and the quality of the team that he

had assembled. If any group could succeed, we concluded, it was

this one.

RMI had talent in the realms of both software and chip design.

It created and refined a set of specifications for the new product in

the process of consulting with potential customers. Production of

the actual chips was contracted to TSMC in Taiwan, the world’s

leading contract chip manufacturer.

This class of semiconductor products is not sold simply as a

chip but as part of a total solution, including the software needed

to integrate the processor into network equipment. Therefore, RMI

started a software development organization in India under the

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide140

leadership of an Indian engineering manager who had returned home

after a ten-year career in the US.

RMI was an international company from the start. In 2009, out

of about 300 employees, about 40 percent were in India and Asia.

The core chip development team stayed in the US, but a large num-

ber of the managers and engineers in the US were born outside the

country.

Furthermore, recognizing that China was likely to be an

important market, RMI had Dr. Sunny Siu build a customer engin-

eering support organization in China. Born in China, Dr. Siu had

earned a Ph.D. from Stanford University in electrical engineering

and had served as assistant professor at MIT – another example of

the phenomenon of Chinese-born technologists moving back to

China and serving as “bridges” between US companies and Chinese

customers.

So, before actually having products to ship in volume, RMI

had prepared the ground to deal with Asian customers to smooth the

sales process.

Product launch

The first of RMI’s new generation of multi-core microprocessors was

completed for customer sampling in 2004, and released for produc-

tion in 2005. It took three years from start to finish. Given the com-

plexity and novelty of the technology, this is record time for getting

a product to market.

Its reception was outstanding. “A new MIPS powerhouse

arrives,” was the headline of the lead article in Microprocessor

Design, the respected industry newsletter, on May 17, 2005. The

article detailed how RMI’s product had outclassed the competition

and was opening new markets. To start with, it delivered a ten-

fold increase in packet processing speed compared to other devices

doing the same job. The article quoted Raza as saying, “The XLR

processor design is my greatest accomplishment to date. It is the

first architecture in my experience that met every architectural

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Product launch 141

goal set for it. Today, this is the highest throughput machine on

the planet.”6

Once RMI’s new product was available, there were some inter-

esting market developments around it. You would have thought that

the first serious customers would be in the US. In fact, they were

in China and Israel. Why? After the failure of a number of US chip

startups in the period following the 2000 market crash, US network

equipment manufacturers became apprehensive of committing new

generations of their equipment to products from private companies

with uncertain economic futures.

At the time RMI was not yet profitable. As a result, although

Juniper in the US and a small network equipment company in Israel

designed the RMI microprocessor into their products, the company’s

success was built on its overseas customers. Sales outside of the US

grew rapidly because new equipment manufacturers in emerging

economies were willing to take a chance on buying from a startup.

The advanced performance of RMI’s new microprocessor gave them

a competitive edge against the dominant vendors in the market, such

as Alcatel and Cisco Systems.

China was home to the most important early customers. These

included Huawei and ZTE, the fast-growing national telecommu-

nications equipment vendors that dominated the Chinese market.

RMI benefited from the explosive growth not only of Internet traffic

but also of wireless traffic, as shown in Figure 5.1. In addition, both

Huawei and ZTE were developing overseas markets for their prod-

ucts. The number of wireless subscribers in China was booming and

local service providers needed equipment to keep up with service

demand. RMI’s microprocessor became the technological underpin-

ning of a new generation of equipment for the local Chinese carriers

such as China Mobile.

As a result of its early focus on Asian customers, RMI had the

right organization on the ground to take advantage of the opportunities

6 K. Krewell, “A new MIPS powerhouse arrives,” Microprocessor Design, May 17, 2005, pp. 1–7.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide142

created by the booming Asian telecommunications market. In 2007

sales in the Asia-Pacific region were $52 million. Sales in the US were

$8 million, and in the rest of the world only $4.6 million.

Transition

In mid-2007 Atiq Raza resigned from the company he had created

and guided to success to return to venture capital management. He

was replaced as CEO by Behrooz Abdi.

Though born in Iran, Abdi received his B.S. degree from

Montana State University and an M.S. degree in electrical engin-

eering from the Georgia Institute of Technology. He joined RMI

from Qualcomm Inc., where he had been Senior Vice President and

General Manager of the chip division. Prior to Qualcomm he had

been at Motorola, where he headed the Radio Products Division.

Abdi joined RMI at just the right time to take it to the next

level as a market leader. His experience with wireless technology

at Qualcomm and Motorola was an enormous asset in this effort

because wireless networks were exactly the market where RMI had

the most to gain. In choosing Abdi, RMI had positioned itself to

adapt to new market realities.

Figure 5.1 Wireless data traffic volume growth by year (figures for 2012 and later are projected). Source: From IBS, private communications, 2011.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Product launch 143

With over 100 patents either issued or pending, the com-

pany had an impressive portfolio of intellectual property to support

expansion into new markets. But it faced a challenge common to all

startup enterprises: the need for more revenue to solidify and extend

its early success. Although its first generation of products had been

launched to great acclaim, RMI had to gain widespread customer

acceptance and build revenues rapidly if it wanted to head off the

competitive pressures it would inevitably face from later entrants

into its market segment.

Abdi’s approach to this challenge was to turn the complex

nature of RMI’s network products into a competitive advantage by

helping its customers design their equipment. In effect, RMI set out

to deliver not only chips but also the software to integrate them into

its customers’ extremely complex communications equipment. This

was particularly important for new Asian equipment builders, who

lacked some of the established engineering skills that US companies

such as Cisco, Juniper, or Lucent had accumulated over the years.

As a result, RMI found itself increasingly in the position of a

system enabler. In addition to expanding its internal chip design and

support organizations, it had to develop a cadre of collaborative soft-

ware vendors if it was to win new customers. It needed these indus-

try partners to provide the enabling software for its customers.

Drawing on his extensive background in communications sys-

tems, Abdi built an ecosystem of over one hundred corporate part-

ners. These organizations developed and sold complementary chips

and software that allowed the company’s customers to build their

products in record time around RMI chips.

By 2008 RMI had over 200 customers and revenues of $79 mil-

lion, an increase of 25 percent from the prior year. It had proven that

it understood the network equipment market, had the technology to

meet its needs, and could work in a global environment. It was in a

good position for future growth – but it needed an infusion of new

capital to fund its expansion. Though successful, it was not yet prof-

itable due to its heavy investment in product development.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide144

Under normal circumstances, the right place to raise this cap-

ital would be the public market. Accordingly Goldman Sachs was

selected to underwrite an IPO on NASDAQ in mid-2008 – just before

the US mortgage crisis triggered a stock market debacle and the deep-

est recession since 1929.

After the 2008 market crash an IPO was clearly impossible.

The company’s board of directors was faced with the task of plan-

ning a different future for RMI.

Weighing options

Deprived of the ability to raise capital in the public markets, RMI’s

directors had two strategic choices.

Keep the company independent; reduce product development costs to • maintain profitability.

Merge RMI with a public company with complementary products and a • similar customer base, in the expectation that the combined companies

would have enough resources to continue their growth.

Staying independent posed the bigger risk. If RMI chose that

course, it would have to cut back on new product development, pos-

sibly mortgaging its future. In a highly competitive industry like

microprocessors, inadequate investment in product development

could be fatal. There were already a number of bigger competitors

racing to overtake RMI’s lead.

If, on the other hand, the company chose the merger route, it

would obviously have to find the right partner. As it happened, in the

course of building RMI’s industry ecosystem, Behrooz Abdi had run

across a compatible company.

RMI merges with NetLogic

NetLogic Microsystems was the world leader in a specialized cat-

egory of semiconductor chips called knowledge-based processors

(KBP). These chips incorporate massive parallel signal processing

with content-addressable memories (CAM), which store the data

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

R MI merges w ith NetLogic 145

needed to move data packets to their destinations. Such specialized

memories can greatly increase a router’s processing speed.

We have already seen that the headers on data packets carry

information to identify packet contents, origin, and ultimate destin-

ation. When the microprocessor in a router analyzes each packet’s

header to determine its destination, it must then look up informa-

tion in its own routing tables to decide how to send the packet on the

next leg of its journey through the network.

CAMs are specially designed to increase the efficiency of the

process of matching a packet’s address information with its destin-

ation path, while meeting network security requirements at the same

time. You can think of the operation as similar to the mail-sorting

procedure in an old-fashioned post office, where the routing informa-

tion for each letter is stored within easy reach of the sorting clerk.

After extensive meetings between the investors and the

managements of RMI and NetLogic, we mutually concluded that

by combining the two companies we would produce a technology

leader with a market position superior to what either company could

achieve on its own.

One avenue of potential growth opened up by this merger,

for example, was the combined company’s ability to develop prod-

ucts that more closely integrated packet processing with NetLogic’s

chips. This would allow customers to increase network performance

while reducing chip costs. In addition, the merger would realize sub-

stantial financial benefits by combining two sales, marketing, and

product development organizations into one, thus reducing admin-

istrative overhead.

We agreed that RMI would exchange its shares for new NetLogic

shares so that shareholders in both companies could benefit from

the new value created by the merger. The merger was completed in

October 2009. RMI’s shareholders ended up with about 20 percent of

the shares of the combined company.

Behrooz Abdi joined NetLogic as executive vice-president

and general manager. In 2010 NetLogic introduced a new family of

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide146

microprocessors that had been in development at RMI. These prod-

ucts extended the market leadership of RMI’s previous offerings,

delivering over four times the processing power at the same price as

its nearest competitors, including entries by Intel. Needless to say,

they were very well received.7

They also required a heavy investment in engineering expert-

ise. You get an idea of the engineering requirements for this kind of

company from its manpower: out of a total staff of 650 at the end of

2010, over 60 percent were engineers.

Despite this technology investment, NetLogic’s financial

performance after the merger did not disappoint its shareholders.

Revenues in 2010 reached $382 million (38 percent in China). Its

market capitalization was about $3 billion in March 2011. This

meant that the shareholders of RMI were credited with a value of

about $600 million for their part of the ownership.

Looking back

RMI successfully challenged the biggest companies in its business

with products that required a unique set of skills. In fact, when the

company was started in 2002, there were only two companies in the

world with the technical skills to design such products. One was

Intel, the industry pioneer and leader in microprocessors. The other

was AMD, whose ability to compete in the market was largely the

result of the initiatives of Atiq Raza, the founder of RMI. However,

neither of these big companies addressed the market that RMI

identified.

But there is a sobering message in this story. Despite having

performed an extraordinary feat of product engineering, its ultimate

destiny as a business was not as a standalone company. The reason is

that an enterprise committed to RMI’s class of highly sophisticated

products requires an enormous investment of resources for continued

7 T. R. Halfhill, “NetLogic broadens XLP family,” Microprocessor Design, July 2010, pp. 1–11.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Look ing back 147

success in the international market. Most startups simply don’t have

the resources to sustain that level of investment unless they have

access to the kind of capital available from public markets.

Still, the story of pre-merger RMI is valuable for what it tells us

about the way entrepreneurs and investors have to think in a global

market. Here are some of the most important points to consider.

Management changes can be healthy. A large part of RMI’s

success must be credited to two outstanding entrepreneurs: Atiq

Raza and Behrooz Abdi. Raza was the visionary leader who got the

breakthrough product designed and launched. Abdi took the com-

pany to the next stage, launching the second-generation product

family while ensuring revenues from the first. His work in build-

ing an ecosystem of complementary software and hardware products

to win customer acceptance was critical to the company’s growth

strategy.

There is no substitute for access to international talent. RMI

attracted outstanding talent from around the world right from the

start. Engineers from India, Pakistan, and China were prominent

both in the Silicon Valley location and in the overseas location.

Their geographical dispersion and their understanding of different

markets, to say nothing of their talent, played a large role in the

penetration of the Asian market.

Startups are often viewed with suspicion by big customers in

the developed economies. The more strategic the component, the

more difficult it is to convince big customers to bet their new prod-

uct lines on an offering from a startup. This was the case with big

companies whose concern about the financial stability of RMI pre-

cluded their purchase of its microprocessors.

A global outlook can lead to willing buyers. RMI’s ability to

address the needs of Asian customers saved the company. Equipment

manufacturers in Asia were more focused on using the most advanced

chips than on the financial viability of their vendors. They wisely

concluded that companies with winning products get financed, and

that their support would make that happen. The company’s ultimate

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use

Speeding voice a nd data tr a ffic wor ldw ide148

success in selling in China and the Asia-Pacific region, however, was

no accident. Right from the start RMI had outstanding marketing and

customer support management located there, and it actively engaged

with potential customers while the products were being designed. It

takes years of effort and very talented people on the ground to gain

the respect of new customers. RMI was prepared.

Industry consolidation is a fact of life. The best products in

the world are no substitute for business scale. This is especially true

for technology companies, which require costly and ongoing invest-

ments in product development. Despite high gross profit margins of

60 percent, RMI was never profitable as an independent company. It

was constantly pouring money into new products, a necessity in an

industry where products are obsolete in less than three years. Under

these circumstances any company with the ambition to remain

independent needs access to large amounts of capital at attractive

valuations.

In the 1990s public markets were very open to financing com-

panies such as RMI. The crashes of 2000 and 2008, however, made

the timing of IPOs problematic. The investors’ decision to merge

RMI with NetLogic recognized that the combined companies would

create more value for their shareholders than if they were independ-

ent. This turned out to be the case – NetLogic remained profitable

after the RMI merger and was valued as a very high performance

growth company – over six times annual revenues. It also spends

nearly 40 percent of its revenues on product development. This is

what it takes to stay a winner in the microprocessor game. NetLogic

was in turn acquired by Broadcom in 2012.

In the next chapter we’ll look at a very different approach

to a similar situation: a company that leveraged IP and partner-

ships, instead of mergers, to reach a scale where it could remain

independent.

EBSCOhost - printed on 1/26/2020 4:19 PM via UNIVERSITY OF THE CUMBERLANDS. All use subject to https://www.ebsco.com/terms-of-use