Even as India looks to boost its semiconductor manufacturing segment, the fabless sector is accelerating growth at a steady pace. A significant player in this segment that has recently received much attention is the Bangalore-based Morphing Machines.
The company's primary expertise lies in developing accelerators tailored for a vast array of applications. This spans from emerging fields such as AI, ML, and generative AI to more advanced applications like 5G, 6G, and even drug discovery in the life sciences domain. These accelerators are designed to be compatible with various processors, whether they're from Intel, ARM, or RISC-V. The advantage of integrating these accelerators is that they can boost performance while simultaneously reducing power consumption by up to tenfold.
The company is in the preliminary stages of deployment. Speaking to Digitimes Asia, Deepak Shapeti, co-founder and CEO of the company, explained that most of their products are in the integration phase, set to be a part of offerings that will launch in the forthcoming 6 to 12 months. Despite this early stage, testing has revealed a couple of significant applications where we've seen remarkable benefits.
Applications and real-world benchmarking
The first application of Morphing Machines' solution is in diffusion algorithms. These are essentially chains of interconnected algorithms, where the output of one directly feeds as the input for the subsequent one. Such algorithms play a pivotal role in various domains.
"They are particularly prevalent in life sciences, in the realm of algorithmic trading in the finance sector, and in modeling disease pathways, especially in understanding complex illnesses like cancer," Shapeti said. "In our preliminary testing phases, carried out in association with partners from the Indian Institute of Science, our processor demonstrated a whopping 45x improvement in its performance. Equally notable was the reduction in power consumption, which plummeted to a mere 7 percent of its original value when stacked up against GPUs."
The company's second application centers on Edge AI. This encompasses AI models that generally pose a challenge for GPUs, leading to the use of accelerators for managing such tasks. Prominent examples include workloads such as AlexNet and MobileNet, which are categorized as neural nets.
"During a real-world benchmarking exercise for one of our clients, who was previously leaning towards a renowned third-party processor from the US, our processor stole the limelight," Shapeti added. "It outperformed their processor in power and performance."
Ensuring cutting-edge presence
The rapidly evolving industry, especially with the demise of Moore's Law, has changed the trajectory of technological advancements. Moore's Law predicted that processors would continually miniaturize, doubling their performance approximately every two years. However, now that the industry hit its physical limits, this law no longer holds true. It is in this context that instead of relying on miniaturization, the focus has shifted towards using accelerators to enhance applications.
"Over the past five years, significant growth and activity have been observed in the accelerator domain," Shapeti said. "Despite this swift evolution, it's essential to understand that the latest isn't always the best in our industry. For instance, a newly founded processor startup might not have an edge over a decade-old one if the older company's foundational strategies are sound."
To ensure continued strength in this segment, Morphing Machines focuses on three primary areas:
Architecture fundamentals: It's vital to have a robust foundational architecture that can anticipate the application needs of tomorrow. Without this, a product can quickly become obsolete.
Reconfigurable hardware: With the unpredictable nature of software advancements outpacing hardware development cycles, the company has adopted a reconfigurable hardware approach. Rather than creating new hardware every time a fresh software application emerges, which would be unsustainable, Morphing Machines has designed its hardware to be adjustable.
"This flexibility ensures we can adapt to new applications efficiently," Shapeti said. "For example, in just five years, AI and ML have seen rapid paradigm shifts, making dedicated processors for these applications impractical. Our solution is to have hardware that can be easily reconfigured to run new applications effectively."
Software support: Finally, hardware is only as good as the software support backing it. Developers prefer not to rewrite their applications continuously.
"Our approach is to allow them to work with what they've already written, making it compatible with our hardware," Shapeti added. "This includes running it at high efficiency, even if it wasn't initially designed for our processor. We achieve this by providing potent abstraction layers, ensuring our relevance across various industry verticals and horizontals."
Conclusion
In the current technological landscape, where advancements are rapid, there's an evident shift towards adaptable and efficient solutions. Morphing Machines has positioned itself within the semiconductor industry by offering custom accelerators.
Balancing between emerging technologies and ensuring compatibility with established ones, the company has responded to the changing dynamics, particularly in light of Moore's Law reaching its limits. They emphasize foundational architecture, reconfigurable hardware, and software support. Fabless startups like Morphing Machines are poised to play a critical role in India's journey toward creating a stronger semiconductor ecosystem.
