The 10 Hottest Semiconductor Startups Of 2021 (So Far)

These startups are taking on semiconductor heavyweights like Intel and Nvidia with new kinds of silicon solutions for compute, storage and networking, many of which are headed for the data center.

New Silicon Solutions For Compute, Storage And Networking

See the latest entry: The 10 Hottest Semiconductor Companies to Watch in 2022

The semiconductor industry may be dominated by juggernauts like Intel, Samsung and Nvidia, but the changing needs of applications, especially those in the data center, have created an opening for a new wave of startups that are creating new kinds of silicon solutions for compute, storage and networking.

These chip startups are working on everything from general-purpose processors that can outperform Intel’s and AMD’s latest to accelerators that can speed up AI and deep learning workloads. And with IDC projecting global semiconductor revenue to continue growing in 2021 after a year of pandemic-induced challenges, these semiconductor startups could benefit from what AMD CEO Lisa Su once called the “insatiable demand for more compute.”

What follows are the 10 hottest semiconductor startups of 2021 so far, and their products range from general-purpose processors and accelerators to network and CPU-offload chips.

For more of the biggest startups, products and news stories of 2021 so far, click here.

Ampere Computing

Top Executive: Renee James, CEO

Ampere Computing is taking on Intel and AMD in the data center with Arm-based CPUs it says can outperform competing processors. The Santa Clara, Calif.-based startup recently disclosed it has landed Tencent, Bytedance, Equinix, CloudFlare and uCLoud as customers in addition to previously disclosed engagements with Microsoft and Oracle. At the same time, the chipmaker is expanding its OEM coverage beyond Gigabyte and Wiwynn to Foxconn and Inspur Group. The company is also starting to design its own custom cores for processors, going beyond its original strategy of using core designs from Arm for its 80-core Altra and 128-core Altra Max CPUs. The company’s CPUs are now publicly available in public instances provided by Oracle Cloud Infrastructure.

Cerebras Systems

Top Executive: Andrew Feldman, CEO

Cerebras Systems is tackling the AI compute market with its massive Wafer Scale Engine 2 chip, which it calls the “largest AI processor ever made.” The Los Altos, Calif.-based startup said its recently announced WSE-2 chip — which comes with 2.6 trillion transistors, 850,000 cores and 40 GB of on-chip memory — is “orders of magnitude larger and more performant than any competing GPU on the market.” The WSE-2 powers the startup’s purpose-built CS-2 AI system, which Cerebras says can deliver “more compute performance at less space and less power than any other system.” The startup’s systems have been adopted by the University of Edinburgh’s EPCC supercomputing center, GlaxoSmithKline, Tokyo Electron Devices as well as the U.S. Department of Energy’s Argonne National Laboratory and Livermore National Laboratory.

EdgeQ

Top Executive: Vinay Ravuri, CEO

EdgeQ is taking on Intel and other players in the 5G infrastructure space with a new, AI-infused modem that can replace multiple components in a base station at the fraction of the cost. The Santa Clara, Calif.-based startup revealed its synonymous “base-station-on-a-chip” earlier this year, promising a 50 percent reduction in total cost of ownership for 5G base stations over competing solutions. The startup showed off the RISC-V-based chip after emerging out of stealth mode last November with $51 million in funding, and it has since added former Qualcomm CEO and Executive Chairman Paul Jacobs and former Qualcomm CTO Matt Grob as advisors. The EdgeQ chip can perform AI functions for edge computing applications while also using AI to improve network capabilities.

Fungible

Top Executive: Pradeep Sindhu, CEO

Fungible wants to make hyperscale data centers available for any organization with its turn-key Fungible Data Center solution that is powered by its namesake data processing unit. With the recently launched data center solution, the Santa Clara, Calif.-based startup claims it can easily slice and dice compute, storage, network and GPU resources on demand while providing “performance, scale and cost efficiencies not even achievable by hyperscalers.” This is all made possible by the Fungible DPU, which can offload various functions from the CPU, including bare metal virtualization, software-defined networking and local storage. The company has raised more than $300 million from investors, including a $200 million Series C round from 2019 that was led by the SoftBank Vision Fund.

Mythic

Top Executive: Mike Henry, CEO

Mythic is taking on Nvidia and other AI chipmakers with its M1076 Analog Matrix Processor, which it says can deliver up to 25 tera operations per second of AI compute while requiring 10 times less power than a typical GPU or system-on-chip solution. The Redwood, Calif.-based startup recently introduced the new M1076 AMP in a variety of form factors for servers and edge devices, saying it can tackle use cases ranging from video analytics to augmented and virtual reality. The launch of the new chip comes after the startup raised a $70 million funding round from Hewlett Packard Enterprise and other investors, bringing its total funding to $165.2 million.

Pliops

Top Executive: Uri Beitler, CEO

Pliops aims to shake up data center storage economics with a hardware accelerator it says can “exponentially” improve cost, performance and endurance for SSD storage. The San Jose, Calif.-based startup earlier this year announced it has raised a $65 million funding round led by Koch Disruptive Technologies, with participation from Intel Capital, Nvidia, Xilinx and Western Digital. The startup says its Pliops Storage Platform has been tested by more than 20 tier-one and enterprise companies, including database software vendor Percona, which said that the “Pliops storage processor is unique in that it is able to increase performance, improve compression and reduce write amplification.”

SambaNova Systems

Top Executive: Rodrigo Liang, CEO

SambaNova Systems is taking a holistic approach to AI computing with hardware, software and services that takes advantage of the startup’s Reconfigurable Dataflow Unit chip. The Palo Alto, Calif.-based startup recently announced it had raised a $676 million funding round led by SoftBank Group that also included the venture arms of Google and Intel. The startup is using the funding to grow market share against Nvidia and other competitors with its subscription-based Dataflow-as-a-Service AI platform, which relies on SambaNova’s RDU-based DataScale system to deliver what it says are “unmatched capabilities and efficiency” for AI.

SiFive

Top Executive: Patrick Little, CEO

SiFive is providing an open-source alternative to Arm’s CPU design business with core designs and custom silicon solutions for AI, high-performance computing and other growing markets based on the open and free RISC-V instruction set architecture. The San Mateo, Calif.-based startup has recently received takeover interest from multiple parties, including Intel, which has reportedly offered $2 billion to acquire the startup. Before the reported takeover interest, SiFive announced that Intel’s new foundry business, Intel Foundry Services, will manufacture processors using SiFive’s processor designs. Last August, the startup raised a $61 million Series E funding round led by SK Hynix, with participation from several other investors, including Western Digital Capital, Qualcomm Ventures and Intel Capital. A month later, the company appointed former Qualcomm executive Patrick Little as its new CEO.

Tachyum

Top Executive: Radoslav Danilak, CEO

Tachyum is taking on Intel, AMD, Nvidia and other silicon compute vendors with what it calls the “world’s first universal processor,” which it says can replace the functions of CPUs, GPUs and other kinds of compute processors while providing higher performance and power efficiency. The Las Vegas-based startup recently announced that customers can now test its Prodigy processor with the company’s FPGA-based emulation system. A four-socket reference design motherboard is expected to be available in the fourth quarter of this year. Tachyum claims that its Prodigy processor can run legacy x86, Arm and RISC-V binaries and outperform Intel’s fastest Xeon processors across data center, AI and high-performance computing workloads while consuming 10 times less power. The startup says Prodigy can also outperform Nvidia’s fastest GPU in AI and HPC workloads.

Xsight Labs

Top Executive: Guy Koren, CEO

Xsight Labs plans to upend the data center switch market with a super-fast, programmable switch it says can meet the power and performance demands of cloud, high-performance computing and AI applications while also providing a flexible and scalable architecture. The Tel Aviv, Israel-based startup announced in March that it had raised a Series D funding round that was backed by several investors, including Intel Capital, Xilinx and Microsoft’s venture fund, M12. The startup launched out of stealth mode last December with the announcement that it is now sampling X1, what it says is the industry’s first switch to offer up to 25.6 terabits per second in speed. The startup says its switch silicon offers these fast speeds at very low power, with less than 300 watts required for the higher end.