Fabless Japanese chip vendor Socionext is haemorrhaging top engineering talent at its Network SoC business unit.

EE Times has learned that at least four people, probably five—described by multiple sources inside and outside Socionext as “a top slice of technical and management people”—are already gone. As of August 1, they’ll be working at Acacia Communications.

The news is significant on two folds. First, Socionext—a fabless vendor born from the merger of troubled LSI design teams from Fujitsu and Panasonic—is still the fledgling venture and must figure out how to survive without key members of a business unit regarded as a pillar of the company.

Second, Acacia, known for its silicon photonics and DSP expertise, is suddenly hip-deep in talent that’s deeply experienced in high-speed DACs and ADCs.

In one fell swoop, Acacia could change the market dynamics of Digital Coherent Optics (DCO) plugged directly to 100G routers and switches.

In mergers and acquisitions, some employee defection is expected. It’s a calculated risk for which corporations prepare. However, it’s unclear if Socionext actually saw this much defection coming this fast, let alone whether it was prepared.

Among the departed are all the very technical staff members at Socionext’s optical networking SoC design team. They played a key role in designing genuinely differentiated technologies, including several generations of custom ASICs, IPs and high-speed ADCs/DACs for long-haul optical network equipment.

These chips are invisible to the outside world. Intrinsic to optical network equipment, they are deeply embedded inside systems designed by companies like Nokia, Huawei and Cisco. Socionext reputedly had the corner on this segment with a global market share of 85% in recent years.

Several sources told EE Times that Socionext’s investors, Development Bank of Japan (DBJ) in particular, regarded the networking SoC design team in the UK as the company’s “crown jewel,” when they architected the merger. When Fujitsu and Panasonic SoC activities merged into a new independent fabless SoC company, the ownership shook out as 40% Fujitsu and 20% Panasonic, with DBJ holding the balance.

In recent months and weeks, Socionext in Japan dispatched executives to stop the UK design team’s brain drain. But a flurry of sudden activity to plug the leak at the last minute made little difference to defectors.

As one source pointed out, the remainder of the mid- to lower-level engineers also have resumes out on the street.

Stiff competition from Broadcom

Socionext is a fabless start-up, only 16-months old. But it’s unusually big, with 2,700 employees and a sprawling product portfolio spanning seven business units.

When it was founded last year, Socionext had plans for an IPO — possibly soon. With revenue estimated at 140-150 billion yen this year, Socionext’s current profit margin is about 10%. The company’s stated goal is to bring it up to 15% by the time the company goes public.

According to sources within Socionext, its Network SoC business unit has been hitting that 15% goal, thanks to its differentiated technology.

EETA socionext 01 Figure 1: Socionext is saddled with a sprawling product portfolio spanning seven business units (Source: Socionext)

But for the optical network chip design team to continue its advance in the long run looks difficult, partly because a management team sent from Japan insists on zero non-recurring engineering cost.

Socionext is already facing a battle from Broadcom, the 800-pound gorilla in the network chip industry. Broadcom is encroaching on the long-haul optical network equipment market by expanding from optical interconnect inside data centres (Ethernet switch to servers, for example) to the data-centre-to-data-centre market.

As Broadcom practices a ‘whatever it costs’ mentality to win design sockets, Socionext faces difficulties. “There were a number of occasions that we couldn’t even bid up for the business,” said a source inside Socionext. “There is a big disconnect between what we see here in the UK and what Socionext management in Japan sees.”

Socionext products occupy the network interfacing with fibre. While network equipment vendors like Nokia and Cisco own their DSP, IP and algorithms, they still need network SoCs and high-speed DACs/DACs from Socionext to complete their systems.

Until recently, there were few off-the-shelf chip vendors in this field. That is, until Broadcom, who has a leg up in the optical interconnect market in data centres, began going after the data-centre-to-data-centre segment—which needs long-haul Dense Wavelength Division Multiplexing (DWDM), an optical multiplexing technology used to increase bandwidth over existing fibre networks.

Big disconnect

Socionext’s design team, armed with 20 ASICs over four generations of process technologies (since it was Fujitsu), is proud of its own technical drive that has resulted in creating new products and cultivating new markets and customers. The team needs a roadmap from 16nm, 10nm and down to 7nm process technologies, said one Socionext source. “Of course no customers have signed onto 7nm process yet but that’s exactly when we start getting ready for it.”

But the response from Japanese managers has been, “Let’s wait for customers’ reactions.” A risk-averse leadership appears solely interested in an established market. One source who believes things will end up very badly for Socionext, said, “It’s like turkeys voting for Christmas.”

Under a spreadsheet-driven management team with “no vision, no strategy and no grasp of the market,” many at Socionext’s Network SoC business unit feel trapped and frustrated.

Next: Behind the exodus of top talents at Socionext