In the past the purview of business intelligence has been high-level: dashboard building, data queries, database management, data access, etc. However, this piecemeal approach cannot be scaled in a transformational way. Data literacy is increasingly becoming a part of what it means to be an engineer or scientist: for example, in addition to a nanotechnologist’s expertise in ultra-small machinery, they must also possess relevant statistical skills to glean meaning from their raw data.
Requiring a statistical mindset of pure and applied scientists comes with some headaches; principally, piercing the mystique around data science and mathematics. This reality led NXP and other large technology conglomerates to a conundrum: how to meaningfully instill statistical thinking across an organization. Breaking inertial budgeting patterns is always challenging, but teams at NXP recognized that investing in an advanced training program would greatly benefit the company long-term.
The solution was to look upward: quantify the business value of applied data science to senior management, and garner support at the very top of the ladder for establishing pockets of advanced statistical enablement that could be scaled globally.
After being greenlit by NXP’s CIO and several VPs, what became the inaugural Citizen Data Science Program (CDSP) ran in early 2019 with the mission to push companywide digital transformation efforts by building out a bank of use cases that quantified impact.
What was once an informal network of individuals advocating for statistical training was now an organized program with an executive mandate.
Today, CDSP is a landmark initiative keeping NXP on the bleeding edge of innovation and talent acquisition in the increasingly competitive semiconductor arena.
How to design digital transformation for 35,000
With C-Suite enthusiasm driving the expansion of CDSP, NXP Enterprise Business Analyst and JMP Enablement Lead Jason Chen designed an upskilling curriculum that evolves every year.
Here’s how it works: individuals interested in the program must bring to the table both a business use case for analytics and their manager’s support. Usually, this is a cohort of around twenty scientists, engineers, and other domain experts who already possess strong statistical knowledge. The 14-week program then combines lectures and applied practice to expand the participants’ toolkits working together in small, cross-functional teams to solve a high-value NXP business challenge. At the end, there is a two-day hackathon-style event in which teams polish and present their use cases.
All told, participants commit roughly 25% of their time to CDSP for the program’s duration—a commitment that clearly demonstrates the value NXP management attaches to the education that CDSP provides, and the dividends it pays.
With CDSP’s domain-agnostic approach, members hail from finance, human resources, process engineering, and beyond. Combining group members with different domain knowledge bases allowed the teams to be more creative and collaborative, while also designing solutions replicable across many NXP departments.
Participants further expand their analytical knowledge and skillset, gain experience working in teams across multiple sites and functions within NXP, and achieve quantifiable business impact through projects with scale-up potential.
This see-think-do approach let CDSP participants learn actively rather than passively: as each member brought a business challenge to the table, there was a sense of urgency in the training. These weren’t just lectures, this was business optimization.
How analytics led to greater project payoff
One example of CDSP’s substantial impact was a project led by Lin Su-Heng, Ph.D., which focused on maximizing yield at a new fabrication facility. NXP had invested $34mn into shifting a project to a new fab but were experiencing 11% lower yields than at the original facility. Each 1% improvement in yield would equate to annual savings of $11.2mn, making the impact clear to NXP and earning Lin the green light.
Utilizing the predictive modeling capabilities of JMP, she and colleagues devised a system to consistently improve the manufacturing yield by using curve analysis. “CDSP taught me how to use the Functional Data Explorer to analyze hundreds, thousands of data curves,” she says. “I can now do that quantitatively, not just eyeball it.”
As a new workflow for addressing yield optimization began to take shape, the team worked to document the impact of the improvement and explore plans to roll newly developed best practices out to other sites. CDSP teammate Francois Bourlon, a Data Expert at NXP, explains that one of the key hurdles to upscaling was simply building awareness about curve analysis. “The topic of dealing with curves is extremely broad,” he says. “My plan is to spread the word: ‘hey, there’s this new way of working and this new concept solution. This isn’t just a script that’s valid for one single activity, it’s a working method.’”
To communicate the potential of adopting new tools like Functional Data Explorer in JMP Pro, Bourlon explains, the team explored ways to understand business impact. Putting efficiencies into the context of salaried experts, Bourlon is optimistic: “Because of better detection levels, [our workflow] is essentially doing the work of one engineer every year who would have normally had to do the analysis by hand. Moreover, there are major gains semi-annually or annually in terms of quality.”
Using JMP to augment decision-making efficiency
In another CDSP project, Santosh Murali, a Device Engineer at NXP, led a project focused on scaffolding a brand-new decisioning system. Device Engineering is accountable for the entire flow of manufacturing from Silicon to shipped product to customers and constantly debugging issues with Design, Process, Equipment, Maintenance, Yield Enhancement, and Product/Test divisions. Through this workflow, millions of data points are generated, stored, and processed in different applications. Therein lies the bottleneck: Murali’s team must identify shifts in manufacturing processes in one platform, then see the raw data in another, then produce solutions in yet another.
Tedium erodes efficiency, and Murali saw an opportunity to propose a business use case to his management that would automate and consolidate the entire data life cycle in JMP. “We wanted to make JMP a decision-making system with interactive visualizations to let it tell us what to do,” he says.
Once implemented, Murali estimates a savings of $650,000 annually. Beyond the quantifiable benefits, he also noted an improvement in “how quickly [NXP] can actually solve problems,” which is “critical in manufacturing.”
The technology to which the consolidated workflow is applied now nears $1bn in revenue annually. But the benefit of designing a solution system is its transferability to other business realms. In the future, Murali hopes to expand this centralized data structure to all NXP technologies, meaning the impact of his CDSP project will ripple far beyond its original domain.
Real Training = Real Results
“A lot of young people are struggling to choose between wanting to be the subject matter expert, or the statistical analyst,” Lin reflects. “My personal experience is that we need both to achieve the highest potential.” This lesson is what many organizations can learn from NXP: to achieve the highest levels of efficiency, data science and domain expertise must be integrated into each team. And the establishment of centers of excellence dispersed across the organization requires an investment in upskilling employees—on top of building connection across departmental boundaries.
At the end of each program, the next step was ensuring that those domain experts who completed CDSP can put their newly cultivated skills to work. A focus on application ensures that NXP maximizes their return on investment, creating a win-win for CDSP participants and for the continued delivery of value to the business. NXP managers have even worked with Human Resources to create new roles where CDSP graduates are better positioned to become statistical thought leaders. This visibility begets more interest in data science, thus repeating the cycle.
Now, a newly established internal data science center aims to connect the dots between individuals and collaboration opportunities. All the effort that NXP has poured into data science—the CDSP, its time commitment, elevating those who complete it, the data science center—ultimately translates to faster innovation cycles and increases the pace at which semiconductor technologies reach the market.
NXP now has over 1,600 JMP Pro users, with growth expected over the coming years as both CDSP and the JMP feature set grow. It’s yet another example of how real enterprise analytics requires real training, and that real training equals real results.