A Holistic-Minded Leader – Hao Zhong: Advancing Artificial Intelligence Revolution with ScaleFlux

Hao Zhong
Hao Zhong

Share on :

Facebook
X
LinkedIn
Pinterest
WhatsApp
Email

Artificial Intelligence (AI) has swept the current times in its sway by leaps and bounds. In fact, the revolutionary promises and transformational possibilities the entire field has brought are unprecedented. The revamp it caused in everything—be it work, play, business, industry, finance, economics, global supply chains, education, healthcare, military, defence, geopolitics, public and private sectors, and personal and professional areas—is mind-boggling.

The astonishing advancements in this exceptional stream of technology are powered by the continuous breakthroughs made by some of the most extraordinary AI leaders, like Hao Zhong, who are revolutionizing the global business horizons.

In all the glitz and glamour of this technological marvel, Hao Zhong never lost his focus on the most crucial element of this evolution—data. He always enjoyed innovation, architecture, and product delivery when it comes to data management. Today, as the Chief Executive Officer (CEO) and Co-founder of ScaleFlux, Inc.,Hao Zhong is leading the storage and memory technology trends, specializing particularly in flash memory-based storage solutions and Compute Express Link (CXL) solutions — from NAND devices, SSD and CXL controllers, to storage systems.

A Constant Spark of Innovation: AI Leadership of 25 Years

With a solid fourteen years of experience with National Instruments, Rensselaer Polytechnic Institute, Broadcom (LSI/Agere), Broadcom (LSI/SandForce), and Fusion-io in engineering and leadership positions, Hao Zhong established ScaleFlux in 2014, in the San Francisco Bay Area. He recalls the time and his own inspiration behind the risk and the opportunity. “The original inspiration came from my experiences in two flash storage pioneers – SandForce and Fusion-io.  Those two companies demonstrated how startups with the right innovations can leapfrog the market.  My co-founders and I sought to recreate that spark of innovation to generate revolutionary improvements for the data pipeline.”

A Holistic Blueprint

Those experiences also gave Hao Zhong a view into how hard it is to scale those innovations efficiently and repeatedly over multiple product generations. Those lessons led him to a holistic blueprint for ScaleFlux:

(1) a highly efficient, scalable ASIC design approach;

(2) a turnkey firmware–software stack (so you’re delivering a whole, validated product—not just a chip);

(3) differentiated IP that moves the TCO needle; and

(4) rigorous qualification and interoperability across flash, DRAM, CPUs/GPUs/TPUs/DPUs—the whole xPU spectrum.

If any one of those pillars is weak, the entire structure wobbles. “That’s been our guiding principle for how we organize the team, run the process, and ship products.”

The Core Mission: Building an Ever-Adaptive Computing Infra

In the face of constantly shifting industry needs and market dynamics, Hao Zhong states that ScaleFlux’s core mission has remained the same – bring new levels of performance and efficiency to the computing infrastructure through innovation in storage and memory technologies – though both the means to accomplish the mission and the definition of computing infrastructure have evolved.

He adds that their initial focus was on computational storage as the means for driving a 10x improvement for data center computing.  Computational storage focused on generating efficiency through reducing data movement and distributing processing closer to where data lived. The concept was solid. The example deployments demonstrated massive improvements in application, system, and cluster-level performance and efficiency. But they were highly customized, proprietary, and difficult to scale to the mass market. “We pivoted our thinking to focus on solutions that could combine tremendous benefits with ease-of-use.” This led them to develop bleeding-edge features and capabilities in standards-based NVMe and CXL solutions. Re-architecting their SoC (system-on-chip) controllers to meet the demands of the AI workloads.

“The explosion in AI over the last few years is what I meant in my reference to the definition of computing infra having evolved,” Hao Zhong explains. AI workloads and the GPUs they run on operate in a fundamentally different manner from traditional CPU-based infra and traditional enterprise applications. Storage and memory must evolve to support this new style of data consumption.

Rearchitecting the Fundamentals

ScaleFlux emphasizes rearchitecture of the SSD and memory controllers for the AI infra rather than turning the crank on traditional architectures. When asked how this fundamentally changes performance and efficiency outcomes for customers, Hao Zhong shares that turning the crank on traditional architectures has had the industry stuck in the pattern of roughly doubling performance and power efficiency at each new generation. “We aim to shift this to a 10x or more improvement with the rearchitecture of the SSD and memory controllers.”

Delivering revolutionary gains instead of modest, evolutionary gains in storage and memory opens the doors for innovation in applications and massive performance and efficiency gains in the AI infrastructure. “One of the biggest challenges with AI is moving data onto and off the GPUs,” he reveals. Traditional storage and memory controllers, which were designed for traditional CPUs, simply will not get the job done. They need to be rearchitected for massive parallelism and finer-grain data accesses, for pooling of resources across processor clusters, and for greater RAS and security. Turning the crank on traditional architectures has resulted in a widening gap between GPUs’ ability to consume data and storage/memory’s ability to deliver the data.

Fixing the Initial Challenges

In translating the founding vision into real products, Hao Zhong recalls the challenges. “Early on, we stayed on FPGAs longer than we should have.” It gave them flexibility, “And let us try new acceleration ideas, but customers couldn’t deploy those at a massive scale. When we did our first ASIC, we stuffed it with nice-to-have features and over-scoped for the market.” The fix was to tighten the market investigation loop, sharpen the target customer and use cases, and bring that discipline directly into their holistic chip-plus-firmware process. You see the outcome in subsequent generations: clearer scope, better alignment, faster learning cycles, and more predictable delivery schedules.

Interlocking All the Strengths: A Four-Pillared Approach

The ScaleFlux team has implemented what they call the “Four Pillars” approach to their development. The pillars are: scalable ASIC design, turnkey firmware/software, differentiated IP, and deep interoperability/qualification. Think of them as the ingredients of a single recipe. Miss the salt, and the dish fails; weaken one pillar and the building shakes. It’s not a checklist, it’s how these parts interlock to reduce tapeouts, accelerate bring-up, and drive system-level outcomes—not just NAND-level optimizations. “That’s also why we embed co-design practices and validate in the full context where customers run: CPUs, GPUs, networks, and real workloads.”

The convergence of AI, cloud, and data-intensive workloads created the right environment and timing for ScaleFlux’s technology. Hao Zhong says that the traditional, “mainstream” enterprise SSDs met or exceeded the data transfer requirements for traditional enterprise applications running on traditional CPUs. CPU-attached memory was reasonably sufficient for the capacity and performance needs of those workloads. The convergence of AI, cloud, and data-intensive workloads fundamentally changes the game, driving the need for innovations in storage and memory to match the new computing paradigm. Combine this technology shift with the seemingly insatiable appetite for AI infra, and you have the perfect environment for the adoption of ScaleFlux technologies. “Our foundational IP puts us in a uniquely strong position to partner and lead in the transition to storage and memory products with extreme performance, density, and efficiency.”

Maintaining the Balance

As performance demands accelerate, to balance maximizing throughput and reliability with sustainability, power efficiency, and overall total cost of ownership, Hao Zhong says they have institutionalized the holistic “ASIC-plus-firmware” approach, invested in differentiated IP, and strengthened interoperability and qualification capabilities. “Equally important, we work with technology leaders earlier in the planning and development process—for example, engaging on accelerated computing for AI—to design in the system context, not just the drive context.” This holistic approach that includes the system and application context integrates sustainability, power efficiency, and TCO into their designs from the planning stage and carries it through development and qualification.

Furthermore, to balance between offering customized, workload-optimized storage solutions and maintaining standardized, interoperable architectures for broad adoption, Hao Zhong  ensures ScaleFlux’s approach is resilient, adaptive, and ever-evolving. First and foremost, the products must help their customers solve problems. Creating products that are standards-based with proven interoperability across platforms makes deployment easier for the customers. Even with innovative features such as in-drive compression, they utilize industry standards. Workload optimizations are typically developed in close collaboration with large-scale customers or are enabled through changing settings in the device. He says, “We learned the lesson with our early generations of products based on proprietary drivers that complexity kills.” Even amazing products need to appear simple for the user.

Building Smart Storage Architecture

To ensure security, reliability, and data integrity, and build customer confidence in “smart” storage architecture, ScaleFlux designs its storage and memory controllers to comply with industry standards for data security and firmware attestation. “We were the first to adopt the Caliptra security blocks in our controllers and continue to lead the way in integrating new iterations of Caliptra and encryption in our products.” For reliability, they have developed advanced error correction capabilities for their controllers to protect against media errors and attacks that target media.

The Strategic Global Expansion

Moreover, to scale globally in terms of partnerships, market strategy, and localization, particularly as a hardware-focused startup operating in international markets, Hao Zhong accepts that this is a rapidly evolving challenge to be sure! “We have scaled out our team globally, leveraging centers of expertise in the US, India, and APAC to recruit the best engineering talent and to provide local support to customers.” They have established both global and regional partnerships with component suppliers and contract manufacturers to provide a reliable, cost-effective supply to customers in each geo.

Nurturing a Culture of Innovation: The Core Leadership Principles

Over more than two decades in data storage and hardware design, there are some core leadership principles that guide Hao Zhong, leveraging which he has cultivated a culture of innovation at ScaleFlux. He adds, “I have a few guiding principles.’ First is to know yourself and continuously seek improvement. These are two sides of the same coin. Knowing what your strengths are and acknowledging areas of weakness is the first step. Acknowledging areas where your knowledge is limited helps prioritize which experts and what content you should seek out first.  Understanding what your strengths are and what areas you simply don’t excel at or don’t have the greatest aptitude for also points you to where you need to hire in expertise.

Second is ‘lead by example.’ You can’t expect your team to put in long hours and hold themselves accountable to excellence, to meet their goals, to strive for continuous improvement, and to persistently fight through obstacles unless you exemplify those traits.

That ties into the third principlefoster a culture of responsibility. Set goals and track progress for yourself and for your team. Take ownership of the wins and the losses.

Finally, as a leader in the semiconductor industry, create products that have a meaningful impact on technology and on society. “Working on products that move us forward gives the team a sense of pride and a sense of purpose to their efforts.”

Tracking the Trends of Tomorrow

In his forecast, Zhong says that the AI infrastructure build-out appears to be set for continued massive growth over the coming years. Semiconductor manufacturing and global power generation limitations are going to constrain just how fast new AI infrastructure is brought online, keeping supply below demand. “We’re seeing components going on allocation not just on the GPUs, but on the storage and memory components to go along with the GPU systems.” This ongoing situation of demand exceeding supply creates a massive opportunity for ScaleFlux, as customers will need new vendors and innovative solutions that will allow them to get more work (and thus more revenue) from their infrastructure.

A Ground-Breaking Advice

Based on his AI leadership journey, in his advice to engineers, developers, or founders aspiring to build deep-tech companies that solve infrastructure-level challenges, Hao Zhong says: Truly ground-breaking solutions to the complex infrastructure-level challenges often come from the startups that are not saddled with legacy products and processes. Startups are easy to doubt. But step by step—by shipping, collaborating, and proving contribution—you change minds. History rewards the innovators with the conviction to bring their concepts to reality and the optimists who engage with new ideas and partners to build what’s next.

Related Articles: