Sehat Sutardja: Marvell Co-Founder and Chiplet Pioneer
Sehat Sutardja is one of the semiconductor industry’s notable engineers and entrepreneurs, best known as a co-founder of Marvell Technology Group and an early advocate of modular chip architectures.
His career spans several major transitions in semiconductor technology. At Marvell, he helped develop digital signal-processing technologies that transformed hard-disk storage. He later became an advocate of multi-die and chiplet-based architectures, anticipating a design approach that has since become increasingly important as traditional monolithic scaling becomes more difficult.
From CMOS-based storage controllers and networking silicon to advanced semiconductor packaging, Sutardja’s career reflects a consistent engineering philosophy: integrate complex systems efficiently while improving scalability, performance, and manufacturing economics.
🚀 Early Life and Passion for Electronics #
Sehat Sutardja was born in 1961 in Jakarta, Indonesia, and developed an interest in electronics at an early age.
As a child, he experimented with electronics projects and studied hobbyist material to understand how circuits and devices worked. One of his early projects involved building a functional Van de Graaff generator with his brother.
That hands-on experimentation eventually developed into a deeper interest in electrical engineering and semiconductor technology.
During his teenage years, Sutardja obtained a radio repair license, giving him practical exposure to electronic systems beyond classroom theory.
He also became interested in major semiconductor companies such as Fairchild Semiconductor, Motorola, and Texas Instruments. Their work helped motivate his decision to pursue advanced engineering education in the United States.
🎓 Academic Journey in the United States #
Sutardja moved to the United States in 1980.
He initially attended the University of San Francisco before transferring to Iowa State University, where he studied electrical engineering.
He subsequently continued his education at the University of California, Berkeley, earning a master’s degree in 1985 and a Ph.D. in Electrical Engineering and Computer Sciences in 1988.
Berkeley was also significant for another reason: Sutardja met Weili Dai, who would later become his wife and business partner.
Their collaboration eventually led to the creation of Marvell and one of the most influential fabless semiconductor companies of its era.
💼 Early Engineering Career #
Before founding Marvell, Sutardja built experience across analog, mixed-signal, and digital semiconductor design.
He worked at companies including Micro Linear, where his work involved digital-to-analog converters and hard-disk-drive-related chips, and Integrated Information Technology, where he worked on digital video compression technology.
These positions exposed him to a combination of analog circuitry, digital signal processing, storage systems, and system-level semiconductor design.
That combination would later become particularly valuable at Marvell, where the company built highly integrated chips for demanding storage and networking applications.
🏢 The Birth of Marvell #
In 1995, Sehat Sutardja co-founded Marvell Technology Group with his wife Weili Dai and his brother Pantas Sutardja.
The company’s early success came from solving difficult problems in data storage.
Revolutionizing Hard-Disk Read Channels #
At the time, hard-disk drives relied heavily on analog signal-processing architectures for reading data from magnetic media.
Marvell pursued a different approach by combining:
- Digital signal processing
- High-speed sampling
- Advanced signal-processing algorithms
- CMOS implementation
- Highly integrated controller architectures
Moving more of the read-channel processing into the digital domain enabled improvements in storage density, performance, and scalability.
Seagate became an important early customer, helping establish Marvell’s position in the storage semiconductor market.
The company’s ability to deliver increasingly sophisticated functionality in compact CMOS devices became a foundation for its subsequent expansion.
Riding the Fabless Semiconductor Transition #
Marvell’s growth also coincided with two important industry transitions.
The first was the expansion of the fabless semiconductor business model, in which chip companies focused on architecture, design, and product development while relying on external foundries for manufacturing.
The second was the industry’s continuing migration toward CMOS-based semiconductor technology.
By designing highly integrated CMOS products and using external manufacturing capacity, Marvell could scale its product portfolio without operating its own large semiconductor fabrication network.
This model eventually allowed the company to expand beyond storage into networking and communications silicon.
🌐 From Storage to Networking #
Marvell’s expertise in high-speed signal processing and integrated semiconductor design naturally extended into networking.
The company developed products including Ethernet switches, transceivers, controllers, and other networking components.
This diversification was strategically important because computing was becoming increasingly interconnected.
As servers, storage systems, and networking infrastructure evolved, the semiconductor industry increasingly required specialized chips capable of moving and processing data at higher speeds while maintaining acceptable power consumption.
Marvell’s design philosophy—high integration combined with specialized processing—was well suited to that transition.
⚠️ Leadership Challenges and Corporate Transition #
Marvell later faced significant corporate challenges.
In 2016, the company became involved in an accounting investigation. Sutardja and Weili Dai were subsequently cleared of wrongdoing, but the period contributed to significant leadership changes.
Sutardja and Dai left their executive positions while Sutardja continued serving as chairman of the board for a period.
The episode marked a transition in his role from operating executive toward broader strategic and technology-focused activities.
🧩 The Chiplet Vision #
One of the most forward-looking aspects of Sutardja’s later career was his advocacy for modular semiconductor architectures.
As conventional semiconductor scaling became increasingly expensive and complex, the industry began exploring ways to construct sophisticated systems from multiple smaller dies rather than relying exclusively on a single massive monolithic chip.
Sutardja was advocating this concept well before chiplets became a mainstream industry strategy.
From MoChi to Modern Chiplets #
At the 2015 IEEE International Solid-State Circuits Conference (ISSCC), Sutardja presented the concept of MoChi, short for modular chip architecture.
The fundamental idea was to divide complex semiconductor systems into multiple functional dies and integrate them within a package.
This approach can offer several potential advantages:
- Greater design flexibility
- Improved manufacturing yield
- Reuse of proven chip blocks
- More efficient product scaling
- Potential cost advantages
- Heterogeneous integration
- Faster development of specialized systems
Instead of treating a processor or accelerator as one indivisible piece of silicon, modular architectures allow designers to combine multiple components according to the requirements of a particular product.
The concept closely parallels the chiplet architectures now being adopted across CPUs, GPUs, AI accelerators, networking processors, and other advanced semiconductor products.
📦 Why Chiplets Became Important #
The rise of chiplets is closely connected to the increasing difficulty of building very large monolithic dies.
As transistor densities increase, large chips become more expensive to manufacture. A defect in a large die can make the entire chip unusable, reducing manufacturing yield.
Smaller dies can potentially improve yield and allow individual components to be manufactured using different process technologies.
A chiplet-based system can therefore combine:
- Advanced compute dies
- Older and more cost-effective process nodes
- High-bandwidth memory interfaces
- I/O dies
- Specialized accelerators
- Networking functions
This creates a more modular approach to semiconductor design.
Sutardja’s early advocacy of modular integration placed him among the engineers and executives who recognized these advantages before chiplets became a dominant industry topic.
🏭 Founding Silicon Box #
In 2021, Sutardja and Weili Dai co-founded Silicon Box, a semiconductor company focused on advanced packaging and chiplet integration.
The company extends the modular semiconductor philosophy into manufacturing and packaging.
Rather than treating packaging as a passive final stage of chip production, advanced chiplet systems require packaging technology capable of connecting multiple dies with extremely high bandwidth and low latency.
This makes advanced packaging increasingly important to overall system performance.
Silicon Box’s focus reflects the broader industry shift toward heterogeneous integration, where semiconductor performance increasingly depends on the interaction between silicon dies, interconnects, substrates, and packaging technologies.
Chiplets Move Beyond a Design Concept #
The significance of Silicon Box is that chiplet technology requires more than an architectural idea.
A practical chiplet ecosystem needs:
- High-density die-to-die interconnects
- Advanced substrates
- Precision assembly
- Thermal management
- High-bandwidth signaling
- Manufacturing consistency
- Packaging processes optimized for multiple dies
In other words, chiplets are simultaneously a chip-design problem and a manufacturing problem.
Sutardja’s move into advanced packaging therefore represents a continuation of his earlier semiconductor philosophy rather than a completely new direction.
🌍 Influence on the Semiconductor Industry #
Sutardja’s influence extends beyond the companies he founded.
His career connects several important semiconductor trends:
- Digital signal processing for storage
- Highly integrated CMOS semiconductor design
- The expansion of the fabless business model
- Networking and data-movement silicon
- Modular multi-die architectures
- Advanced chiplet packaging
These areas are tightly connected to the architecture of modern computing.
Storage systems require sophisticated signal processing. Networking systems require high-speed data movement. AI accelerators require enormous compute and memory bandwidth. Advanced processors increasingly rely on multiple dies and sophisticated packaging.
The common thread is system-level integration.
Engineering over publicity #
Sutardja has generally been recognized within the semiconductor industry for his technical orientation and long-term approach to engineering.
His work illustrates how semiconductor innovation often develops over decades.
Ideas that initially appear impractical can become important when manufacturing economics, process technology, and system requirements eventually align.
Chiplets are a strong example.
The concept of breaking large semiconductor systems into modular components existed before the industry had the packaging technology and economic incentives required for widespread adoption. As monolithic scaling became increasingly expensive, those earlier ideas became substantially more relevant.
🧠 A Career Spanning Multiple Semiconductor Eras #
Sutardja’s career mirrors several major stages of semiconductor evolution.
He began with hands-on electronics experimentation before moving into formal electrical-engineering research. He then entered Silicon Valley’s semiconductor industry, working on analog and digital technologies before co-founding a company that became a major force in storage and networking.
At Marvell, the emphasis was on integrating more functionality into CMOS silicon.
Later, his attention shifted toward the opposite architectural direction: breaking increasingly complex systems into modular components that could be integrated at the package level.
At first glance, these approaches may appear contradictory.
They are actually two responses to the same engineering challenge: how to efficiently integrate increasingly complex computing systems.
The difference is where the integration happens.
Earlier generations emphasized putting more functionality onto a single die. Modern chiplet architectures increasingly combine multiple dies within one package.
🧾 Legacy of a Semiconductor Pioneer #
Sehat Sutardja’s legacy extends beyond Marvell’s financial or commercial success.
His early work helped advance digital storage technology during a period when hard-disk capacity was increasing rapidly. His leadership contributed to Marvell’s expansion from storage into networking and other high-performance semiconductor markets.
His later advocacy for modular chip architectures anticipated a major direction in modern semiconductor design.
Through Silicon Box, that vision has also expanded into advanced packaging and chiplet manufacturing.
The semiconductor industry is now increasingly moving toward heterogeneous integration, where performance depends not only on transistor density but also on how effectively multiple dies, memory technologies, interconnects, and packaging systems work together.
That trajectory makes Sutardja’s career particularly relevant to understanding where semiconductor architecture is heading.
His story is ultimately one of integration evolving from the transistor level to the system and package level—from digital storage controllers and networking silicon to the chiplet-based architectures increasingly shaping next-generation computing.