SoftBank chairman Masayoshi Son. Credit: Digitimes
SoftBank chairman and CEO Masayoshi Son raised the possibility of a strategic alliance between Arm and Samsung on September 22, according to several media outlets, as falling technology stocks worldwide complicate SoftBank’s ongoing plan to publicly list Arm in 2023. According to Nikkei AsiaSon announced plans to visit South Korea for talks with Samsung regarding the strategic alliance, in a bid to boost the chip design company’s valuation.
SoftBank acquired Arm for US$32 billion in 2016, and after the US$40 billion acquisition attempt from Nvidia failed in February 2022, Arm’s IPO became a priority. Nikkei Asia reported that Son had been thinking about what to do with Arm “almost everyday.” In the first quarter of fiscal 2022, SoftBank posted a record JPY3.16 trillion (US$23.5 billion) loss. At the same time, however, Arm posted a record revenue of US$719 million in the first-quarter of fiscal 2022, with its quarterly royalty revenue reaching beyond US$400 million for the first time. Previous reports indicated that SoftBank aimed to float Arm before March 2023, pursuing a valuation of at least US$60 billion.
A Samsung executive told Financial Times that “a strategic alliance is a vague and broad term,” and that the South Korean company would have to “consider it on a general basis” if SoftBank offered to sell Arm to Samsung. The South Korean memory giant SK Hynix also considered Arm as a potential acquisition target during a March shareholders meeting, as reported by Nikkei Asia. On multiple occasions in the past two years, industry leaders like Samsung, SK Hynix, Intel and Qualcomm have reportedly raised interests to acquire Arm via a consortium, but there has been no concrete plan in sight, especially after Arm filed a lawsuit against Qualcomm in September 2022.
Arm’s IPO has been further complicated as London launched a renewed attempt to persuade SoftBank to pursue a dual listing. Former British prime minister Boris Johnson had been in talks with SoftBank in an attempt to have Arm listed on the London Stock Exchange, but the talks were halted following Johnson’s departure from the government. Elizabeth Truss, the new British prime minister, recently re-started the talks with SoftBank. The plan to list Arm on NASDAQ is not only complicated by turbulence in the stock market, however. Personnel involved in the talks revealed that the recent US government’s decision to block Nvidia from shipping advanced chips to China had led SoftBank to take US regulatory risks into consideration, according to Financial Times.
Arm’s potential partnership with Samsung is not merely financially motivated, nevertheless. “Potential collaboration with Samsung includes near-memory compute for AI, and Samsung could provide the memory that fits the required memory architecture,” a senior Arm executive revealed to DIGITIMES Asia. Indeed, the growing need to handle AI workloads is showing the limits of the classical Von Neumann architecture – in which data and instructions are both stored in the same memory accessed via a single bus by the CPU – hitherto underpinning general-purpose computing. Despite its programmable nature suitable for machine learning, the energy spent on moving data between processors and memories has become a drag on efficiency. To address the memory bottlenecks, near-memory and in-memory computer architectures have become two alternatives gaining traction, as they shorten the distance between memory and logic units.
In fact, when it comes to in-memory computing, Samsung claimed in January 2022 that it had successfully developed the world’s first in-memory computing based on a magnetoresistive RAM (MRAM) array chip that would see data storage and data computing performed in the same location. According to the company, the chip achieved an accuracy of 98% in classification of hand-written digits and a 93% accuracy in detecting faces from scenes. The research was led by Samsung Advanced Institute of Technology in collaboration with Samsung Electronics Foundry Business and Semiconductor R&D Center.