Chip startup SiFive has launched four CPU cores optimized for running artificial intelligence models. These new cores, based on the open-source RISC-V architecture, add vector extension capabilities that allow for more efficient parallel processing of multiple data points, significantly accelerating AI model computations. The X160 and X180 are the flagship products, featuring vector processing capabilities that accelerate convolution operations, suitable for industrial devices, consumer electronics, and data centers. The company expects customers to begin producing chips based on the new core designs in the second quarter of 2026.
Chip startup SiFive Inc. today announced four new central processing unit cores optimized for running artificial intelligence models.
Headquartered in Santa Clara, California, SiFive was valued at $2.5 billion after its latest funding round in 2022. The company develops CPU designs based on the open-source RISC-V instruction set architecture (ISA). The ISA defines the underlying set of computational operations used by chips to process data.
SiFive’s new cores extend the functionality of RISC-V by adding vector extensions. These components enable the chips to process multiple data points more efficiently at the same time. The parallel computing capability can significantly accelerate large language models, which use a large number of relatively simple computations to process data.
The new product line is led by the X160 and X180. Both designs include vector processing capabilities that accelerate convolution operations, which are computations used by computer vision models to process images. These features can also accelerate certain data filtering and transformation tasks.
The X160 and X180 can be configured to run real-time operating systems. This type of operating system ensures that computations are always completed within a predetermined time, which is crucial for certain systems. For example, manufacturers may want to ensure that factory sensors always generate an alert within three seconds of detecting a device failure.
The X160 is equipped with up to 200 kilobytes of cache and 2 megabytes of memory. In addition to industrial devices, SiFive also believes this chip can be used in consumer devices such as fitness trackers. Furthermore, the X160 can be installed in systems containing multiple AI accelerators to manage chips and prevent firmware tampering.
The X180 is 10% faster than the X160. It can also store more data through two built-in caches, with a total capacity exceeding 4 megabytes. According to SiFive, this chip is suitable for training large language models and powering certain types of data center devices.
The company also announced two additional cores today, which are upgrades of existing designs. The X280 targets consumer devices such as augmented reality headsets, while the X390 can also power automotive and infrastructure systems. The vector processing speed of the latter has quadrupled.
According to SiFive, its engineers have enhanced these two designs with a new coprocessor interface. This technology will make it easier to integrate the cores into system-on-chip designs that also include AI accelerators. Additionally, the company has upgraded the memory subsystems of the X280 and X390 to reduce latency.
SiFive expects customers to begin producing chips based on its new core designs in the second quarter of 2026.
Q&A
Q1: What are the features of SiFive’s RISC-V processor cores?
A: SiFive develops CPU designs based on the open-source RISC-V instruction set architecture, and the new cores enhance functionality by adding vector extensions, allowing the chips to process multiple data points more efficiently, significantly accelerating the operation of artificial intelligence models.
Q2: What are the main applications of the X160 and X180 processor cores?
A: The X160 is equipped with 200 kilobytes of cache and 2 megabytes of memory, suitable for industrial devices, consumer devices such as fitness trackers, and managing multi-AI accelerator systems. The X180 is 10% faster than the X160, with a cache capacity exceeding 4 megabytes, primarily used for training AI models and powering data center devices.
Q3: When will SiFive be able to mass-produce these new processor cores?
A: SiFive expects customers to begin producing chips based on its new core designs in the second quarter of 2026. Currently, these designs are still in the release phase and require time for system integration and manufacturing preparation.