Silicon Box launches $2B semiconductor packaging fab in Singapore

Semiconductor startup Silicon Box has launched a $2 billion superior semiconductor packaging fab in Singapore.

It’s a really costly wager on a brand new chip packaging know-how, nevertheless it comes from seasoned entrepreneurs/buyers Weili Dai, Sehat Sutardja (each founders of Marvell) and CEO Byung Joon “BJ” Han, a chip packaging veteran.

The facility on the Tampines Wafer Fab Park goals to revolutionize chip manufacturing, develop native capabilities, and enhance Singapore’s place as a worldwide hub for semiconductor manufacturing. The 73,000-square-meter manufacturing unit is equal to fifteen soccer fields. It ought to have 1,200 to 1,400 individuals when it’s full, and it has the help of the Singapore Economic Development Board (EDB).

In an interview with VentureBeat, Dai, Sutardja and Han stated the corporate — which they based in 2021 — discovered a gap because of the most recent developments in chip manufacturing and design generally known as “chiplets,” a time period coined by AMD chief know-how officer Mark Papermaster.

As chip miniaturization slows down due the restrictions of physics, Sutardja foresaw way back to a speech he gave in 2015 {that a} new sort of chip packaging would emerge, the place a number of chips could be related collectively as one in a single package deal.

“I was proposing this technology in the ISSCC talk in 2015 for the purpose of building your smartphones, your laptops, everything,” Sutardja stated.

He didn’t foresee on the time that AI would require the know-how, however that day has come in a really large means.

The rise of chiplets

Those chips and chiplets could be wired along with 1000’s of steel connectors. Lately, large chip firms similar to Advanced Micro Devices have endorsed the usage of chiplets in each central processing items (CPUs) and graphics processing items (GPUs).

Chiplets are a technique to hold advancing chip design in mild of the slowdown of Moore’s Law, formulated in 1965 by Intel chairman emeritus Gordon Moore. He predicted know-how would progress in order that chip makers would have the ability to double the variety of parts on the same-size chip each couple of years. And that might enhance efficiency (as dense chips shorten the space that electrons must journey) and cut back prices as effectively.

But Moore’s regulation has both halted (as prompt by Nvidia CEO Jensen Huang) or simply slowed down (as famous by Intel CEO Pat Gelsinger). Part of the reason being that the circuitry simply can’t be miniaturized so simply now. The layers for the bodily buildings are just some atoms thick now, and the width between circuits is round 5 nanometers, or 5 billionth of a meter.

Back in 2015, Sutardja noticed that the price of making chips was going up exponentially.

AMD engineers have stated complete chip trade is probably going shifting to chiplets as a result of the manufacturing enhancements which have pushed the chip trade for many years are diminishing.

AMD tried out chiplets first with its Ryzen processors and now it’s including them to Radeon graphics. With the most recent Radeon chips, it focused the 5-nanometer manufacturing course of for the graphics chip and 6-nanometer manufacturing for six accompany chiplets that present fast-access cache reminiscence.

All of these chips are packaged collectively in the identical module in order that it’s simpler to create quick connections between the processing and reminiscence parts. Su stated the designed helped obtain the 54% increased efficiency per watt, 18% increased frequency, 2.7 instances peak bandwidth at 61 teraflops, and two instances directions per clock in comparison with the prior era.

“The concept of chiplets – building larger systems on chips with modular components produced at scale – allows us to enable chip designers to focus on optimizing for performance and low power consumption in large chip designs, at a cost that is not prohibitive,” Sutardja stated. “Silicon Box’s proprietary fabrication method, sets a new standard for design flexibility and electrical performance at low cost. This agility in semiconductor design cycles enables the industry to take advantage of the chiplet concept and bring designs that double computing performance, at up to four times lower costs for graphical processors and high-performance computing chips, and up to half the cost for more widely consumed mobile processors.”

Silicon Box’s answer

These are large desires for Silicon Box, which has 70 individuals and is hiring quickly.

Dai stated that Silicon Box’s proprietary fabrication technique units a brand new commonplace for design flexibility and electrical efficiency at low value.

“Somebody has to put the Lego pieces together,” Sutardja stated. “To make AI tech work at high speed and low latency, we really need some new technology.”

Silicon Box is ready to make large chips which can be extra like panels — just like the panels for flat panel shows — that may be a lot as 5 instances as massive as a typical silicon wafer. They could have a number of chiplets on them, making them 9 instances extra environment friendly by utilizing the shortest attainable connections throughout the chips. The aim is to construct 9,000 panels per week.

Other tech firms like Cerebras have moved to creating large chips out of full wafers, however Silicon Box is making use of such concepts to the chip packaging problem.

While this Silicon Box manufacturing unit isn’t costly to construct as a $10 billion to $20 billion wafer fabrication manufacturing unit, it nonetheless has subtle know-how similar to a clear room and costly packaging tools.

“It’s expensive as we have to build scale and huge capacity,” Sutardja stated. “For the AI applications alone, the requirement is many times what our factory can build.”

Sutardja, chairman of Silicon Box, believes this agility in semiconductor design cycles permits the trade to make the most of the chiplet idea and convey designs that double computing efficiency, at as much as 4 instances decrease prices for graphical processors and high-performance computing chips, and as much as half the associated fee for extra extensively consumed cellular processors.

Rivals similar to Intel have moved to chiplets as effectively, however Intel doesn’t have the identical tech that Silicon Box has, Han stated.

“We need to get to scale,” Han stated. “We’ll have investors. The tech is successfully developed. The concept is proven. And our cash flow will support the $2 billion cost.”

The firm’s interconnection know-how goals to resolve the distinctive problem for chiplets, that are important to energy emergent applied sciences throughout areas similar to synthetic intelligence (AI), information centres, electrical autos (EVs), cellular, and wearables, Dai stated. She stated Sutardja and Han had been eager about the packaging issues for many years.

Silicon Box has fashioned partnerships globally with trade leaders to allow semiconductor options required to energy next-generation purposes.

The packaging facility can even present coaching alternatives for graduates from institutes of upper studying to contribute to the event of radical new applied sciences. Silicon Box will function with economically sound processes to reduce the damaging results on the surroundings, guaranteeing that its interconnected chiplets are energy-efficient at scale. A totally contained wastewater remedy facility permits 50% of post-manufacturing wastewater to be recycled and reused.

“Silicon Box’s decision to set up its first manufacturing and R&D facility in Singapore is testament to our competitiveness as a critical global node for semiconductors, and a vote of confidence in the long-term growth prospects of the sector in Singapore,” stated EDB’s Chairman Png Cheong Boon, in a press release.

Han stated that the corporate doesn’t want $2 billion on the outset, as it should make purchases of apparatus over time that fill out the manufacturing unit because it reaches full manufacturing.

“Silicon Box is well poised to solve the unique challenge for chiplets, which are essential to power emergent technologies,” Han stated. “Our team of experts with over 30 years of experience, critical ecosystem of partners, and proprietary interconnection technology will shorten the design cycle of chiplets, lower new device costs, reduce power consumption and enable faster-time-to-market for industry partners involved in areas such as artificial intelligence, data centers, electronic vehicles, mobile, and wearables.”

Since leaving Marvell (after an accounting scandal) seven years in the past, Dai and Sutardja have develop into energetic buyers in startups similar to MeetKai, a metaverse startup. MeetKai helped create the Silicon Box facility with a digital twin of the positioning.

“At MeetKai, we are thrilled to have partnered with Silicon Box and have barely touched the surface of what promises to be very fertile ground for enterprise and industrial use cases in the metaverse,” stated James Kaplan, CEO of MeetKai, in a press release. “As a disruptive leader in the metaverse space, it gives us great pride to work with a company moving so disruptively in the semiconductor space. Our heavy usage of AI has given us a first hand look at the needs for mass disruption in the semiconductor space. By automating training and monitoring capabilities, we hope to set the groundwork for Silicon Box to scale to many more factories even faster.”

Silicon Box has fashioned partnerships globally with trade leaders to allow semiconductor options required to energy subsequent era purposes.

“What the users of tomorrow will expect when it comes to their end experience, whether it is on a mobile device, electric vehicle, or an end application utilizing generative AI, requires fundamentally different approaches to semiconductor manufacturing such as the advanced panel level packaging technology in which Silicon Box stands at the forefront,” stated Dai.

She believes that is the primary of many services.

“The Singapore factory is just the beginning, the flagship, but the whole world is moving forward with advanced packaging,” Dai stated.

VentureBeat’s mission is to be a digital city sq. for technical decision-makers to achieve data about transformative enterprise know-how and transact. Discover our Briefings.